Simulations of the WFIRST Supernova Survey and Forecasts of Cosmological Constraints
R. Hounsell, D. Scolnic, R. J. Foley, R. Kessler, V. Miranda, A. Avelino, R. C. Bohlin, A. V. Filippenko, J. Frieman, S. W. Jha, P. L. Kelly, R. P. Kirshner, K. Mandel, A. Rest, A. G. Riess, S. A. Rodney, L. Strolger
MMNRAS , 1–29 (2016) Preprint 8 February 2017 Compiled using MNRAS L A TEX style file v3.0
Simulations of the
WFIRST
Supernova Survey andForecasts of Cosmological Constraints
R. Hounsell , , (cid:63) D. Scolnic , R. J. Foley , R. Kessler , V. Miranda , A. Avelino R. C. Bohlin , A. V. Filippenko , J. Frieman , , S. W. Jha , P. L. Kelly ,R. P. Kirshner , , K. Mandel , A. Rest , A. G. Riess , , S. A. Rodney , L. Strolger Department of Astronomy and Astrophysics, University of California Santa Cruz, 1156 High St., Santa Cruz, CA 95064, USA Department of Astronomy, University of Illinois Urbana Champaign, 1002 W Green St., Urbana, IL, 61801, USA Kavli Institute for Cosmological Physics at the University of Chicago, 5620 S Ellis Ave., Chicago, IL 60637, USA University of Pennsylvania Department of Physics & Astronomy, 209 South 33rd St., Philadelphia, PA, 19104-6396 USA Harvard-Smithsonian Center for Astrophysics, 60 Garden Street,Cambridge MA 02138, USA Space Telescope Science Institute, 3700 San Martin Dr., Baltimore, MD 21218, USA Department of Astronomy, University of California, Berkeley, CA 94720, USA Fermi National Accelerator Laboratory, P. O. Box 500, Batavia, IL 60510 Department of Physics and Astronomy, Rutgers, the State University of New Jersey,136 Frelinghuysen Rd., Piscataway, NJ 08854, USA Department of Physics and Astronomy, University of South Carolina, 712 Main St., Columbia, SC 29208, USA Gordon and Betty Moore Foundation, 1661 Page Mill Road, Palo Alto, CA 94304, USA Department of Physics and Astronomy, The Johns Hopkins University, 3400N. Charles St., Baltimore, MD 21218, USA
Accepted XXX. Received YYY; in original form ZZZ
ABSTRACT
The
Wide Field InfraRed Survey Telescope ( WFIRST ) was the highest rankedlarge space-based mission of the 2010
New Worlds, New Horizons decadal survey. Itis now a NASA mission in formulation with a planned launch in the mid-2020’s. Aprimary mission objective is to precisely constrain the nature of dark energy throughmultiple probes, including Type Ia supernovae. Here, we present the first realisticsimulations of the
WFIRST
SN survey based on current hardware specifications andusing open-source tools. We simulate SN light curves and spectra as viewed by the
WFIRST wide-field channel (WFC) imager and integral field channel (IFC) spec-trometer, respectively. We examine 11 survey strategies with different time allocationsbetween the WFC and IFC, two of which are based upon the strategy described bythe
WFIRST
Science Definition Team, which measures SN distances exclusively fromIFC data. We apply selection criteria and analysis methods based on recent SN cos-mological analyses. We propagate statistical and, crucially, systematic uncertaintiesto predict the dark energy task force figure of merit (DETF FoM) for each strategy.The increase in FoM values with SN search area is limited by the overhead times foreach exposure, and the dependence of the FoM on the maximum redshift is limited bythe parameterisation of dark energy. For IFC-focused strategies the largest individualsystematic uncertainty is the wavelength-dependent calibration uncertainty, whereasfor WFC-focused strategies, it is the intrinsic scatter uncertainty. We consider theimpact of potential reductions to each systematic uncertainty before launch, resultingin a range of FoMs for each strategy. We find that the best IFC-focused and WFC-exclusive strategies have comparable FoM values. Even without improvements to othercosmological probes, the
WFIRST
SN survey has the potential to increase the FoMby more than an order of magnitude from the current values. Although the surveystrategies presented here have not been fully optimised, these initial investigations arean important step in the development of the final hardware design and implementationof the
WFIRST mission.
Key words: surveys – space vehicles: instruments – (stars:) supernovae: general –(cosmology:) dark energy – techniques: imaging spectroscopy (cid:63)
E-mail: [email protected] (cid:13) a r X i v : . [ a s t r o - ph . I M ] F e b R. Hounsell et al.
The
Wide-Field InfraRed Space Telescope ( WFIRST )is a NASA mission that will constrain the nature of darkenergy through multiple probes. It was the top large space-based mission from
New Worlds, New Horizons , the last USastronomy and astrophysics decadal survey (National Re-search Council 2010). As its name suggests,
WFIRST is opti-mised for near-infrared observations and it possesses a largefield of view (FoV). The mission is in formulation at NASA,and several concepts have been suggested so far (Spergelet al. 2015). The current design utilizes a telescope that wasdonated in 2012 by the National Reconnaissance Office. Theaperture of the telescope is the same as the
Hubble SpaceTelescope ( HST ), with both having 2.37-m primary mirrors.Two main instruments are proposed for
WFIRST : a coron-agraph, which will be used for exoplanet and planetary diskstudies, and a wide-field instrument which will be used toprobe dark energy models. The wide-field instrument is it-self composed of a wide-field channel (WFC) imager andintegral-field channel (IFC) spectrometer.Two major
WFIRST goals are to measure the cosmolog-ical growth of the Universe as well as to probe its geometryon large scales. To achieve these two milestones,
WFIRST will conduct multiple observational programs, one of which isa supernova (SN) survey. Type Ia supernovae (SNe Ia) haveplayed a critical role in the discovery of the acceleration ofthe Universe’s expansion (Riess et al. 1998; Perlmutter et al.1999). Recent analyses using multiple cosmological probes(e.g., Betoule et al. 2014; Planck Collaboration et al. 2016;Alam et al. 2016) are all consistent with a Universe that isgeometrically flat, and that is filled primarily with dark en-ergy that behaves like a cosmological constant and cold darkmatter (the ΛCDM model; e.g., Peebles 1984; Efstathiouet al. 1990; Frieman et al. 2008a). There remain however,theoretical arguments for alternatives to the cosmologicalconstant (e.g. Weinberg 1989; Frieman et al. 2008a), whichcan serve as additional motivation for a new generation ofexperiments.The dark energy equation of state can be used to distin-guish between many alternative explanations for the accel-erated expansion of the Universe (e.g., see Joyce et al. 2016,for a review of dark energy and modified gravity), and it isparameterised as, P = wρc , (1)where P and ρ are the dark energy pressure and energy den-sity, respectively, and w is its equation-of-state parameter.In some models, the dark-energy equation of state evolveswith time, and one common parameterisation (proposed byChevallier & Polarski 2001; Linder 2003), that we adopt inthis work is, w = w + (1 − a ) w a , (2)where a = (1 + z ) − is the scale factor of the Universe, w is the current value of the equation-of-state parameter, and w a parameterises its evolution. For a cosmological constant, w ≡ − w a ≡ w , the Dark EnergyTask Force (DETF; Albrecht et al. 2006) suggested the useof a Figure of Merit (FoM): defined as the inverse of thearea enclosed within the 95% confidence contour in the w − w a plane, to compare the capabilities of different surveysin constraining the dark-energy equation of state. Currentconstraints on ( w ; w a ) are( w ; w a ) = ( − . ± . − . ± . , (3)which correspond to a FoM of 32.6 in Alam et al. (2016) (seealso Betoule et al. 2014, where FoM = 31.3). This FoM valueincludes the use of SNe, without SNe Alam et al. (2016)obtains a FoM of 22.9.Understanding the nature of the largest component ofthe Universe is an important goal and one in which the com-munity has invested significant resources. The DETF iden-tified different “stages” of dark energy experiments startingwith initial studies, Stage 1, and progressing towards Stage 4surveys in the mid 2020’s. Stage 3 experiments are currentlyunderway (e.g., the Dark Energy Survey DES Collaboration2005) and are expected to increase the FoM by a factor3 to 5 over Stage 2 experiments. Going forward, all darkenergy surveys are likely to be limited by systematic uncer-tainties. WFIRST is a Stage 4 experiment, and it is designedto reach a factor of ten gain over Stage 2 experiments (i.e.,FoM (cid:38) . < z < .
1) SN Ia data have been obtained bygroups/surveys such as the Center for Astrophysics 1-4 (CfARiess et al. 1999; Jha et al. 2006; Hicken et al. 2009a,b, 2012),the Carnegie Supernova Project (CSP, Contreras et al. 2010;Folatelli et al. 2010; Stritzinger et al. 2011) the Lick Ob-servatory Supernova Search (LOSS, Ganeshalingam et al.2013) and the Foundation SN survey (Foley et al., in prep).SNe Ia at higher redshifts (1 . < z < .
1) have been exam-ined by surveys including ESSENCE (Miknaitis et al. 2007;Wood-Vasey et al. 2007; Narayan et al. 2016), the Super-Nova Legacy Survey (SNLS, Conley et al. 2011; Sullivanet al. 2011), Sloan Digital Sky Survey (SDSS, Frieman et al.2008b; Kessler et al. 2009b; Sako et al. 2014a) and Pan-STARRS1 (PS1, Rest et al. 2014; Scolnic et al. 2014b). Todate, some of the highest redshift ( z > .
0) SNe Ia havebeen observed by the Supernova Cosmology Project (SCP,Suzuki et al. 2012), GOODS (Riess et al. 2007), the CosmicAssembly Near-infrared Deep Extragalactic Legacy Survey(CANDELS, Rodney et al. 2014) and the Dark Energy Sur-veys SN program (DES-SN, Bernstein et al. 2012). Thesesurveys form our current state-of-the-art cosmology sample,consisting of over 1000 spectroscopically confirmed SNe Ia,and extending the Hubble diagram out to z ∼ WFIRST
Science Definition Team (SDT)outlined a baseline 6-year mission, including a two-year SNsurvey, corresponding to 6-months of “on-sky” time (Spergelet al. 2015). The focus of our paper is to deepen the discus-sion on this survey, and progress towards a more optimised
WFIRST
SN strategy. Based on a state-of-the-art analysiswe investigate the impact of systematic uncertainties on thedark energy FoM. See
MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey A greater understanding of systematic uncertainties andtheir effects is obtained by accurately simulating the surveywith sophisticated analysis software such as the SuperNovaANAlysis (SNANA; Kessler et al. 2009a) package. SNANAis built to create highly accurate simulations of SN surveys,and model the impact of systematic uncertainties. It hasbeen used in several cosmology analyses (Sako et al. 2014b;Rodney et al. 2014; Rest et al. 2014; Betoule et al. 2014) toperform light-curve fitting and predict bias corrections fora variety of surveys including low- z , SDSS, PS1, SNLS, and HST . It is routinely updated with the most current tech-niques for simulations and analysis. Using SNANA in ad-dition to several other open-source tools, we have designedand evaluated various
WFIRST
SN survey strategies, cre-ating detailed simulations and conducting a thorough in-vestigation of uncertainties. Our simulations are the first oftheir kind for the
WFIRST mission and allow us to predictand compare the potential scientific impact of each strategy.Furthermore, our work acts as a reference for future simu-lations and provides a guide for the ongoing planning of the
WFIRST mission.We structure this paper as follows. We describe
WFIRST and its instruments in Section 2. Section 3 presentsan outline of the SDT SN survey strategy, while Section 4provides a comprehensive description on how we applied alltools to create the various SN simulations. Additional surveystrategies as well as analyses of those strategies examined arepresented in Section 5. We explore different assumptions forvarious systematic uncertainties and outline their impact onthe FoM measured by
WFIRST simulated SN surveys in Sec-tion 6. Section 7 compares the simulated survey strategiesdescribed in this work, with Section 8 providing a discussionon future considerations for the optimisation of the
WFIRST
SN survey. Finally Section 9 presents our conclusions. WFIRST
HARDWARE:
Planned for launched in the mid 2020’s
WFIRST is ex-pected to be placed into an L2 orbit (1.5 million km awayfrom the Earth at the second Lagrange point), where it willreside for the duration of its 6 year mission. Analogous to
HST , WFIRST consists of a primary mirror that is approx-imately 2.37 meters in diameter. Light from the primary isreflected to the on-axis secondary mirror, which then feedsinto the paths of its various instruments. The design of thetelescope is not yet finalised, however current plans call forboth a wide field instrument (WFI) and a coronagraph .For the purpose of this paper we focus on the WFI only.When preparing our simulations we used the best-availableWFI hardware specifications; these were taken from the May25 th , and an operational temperature of 260 K is assumed. The WFI has two optical channels: the first is a WideField Channel (WFC), the second an Integral Field Channel For more information on the coronagraph see http://wfirst.gsfc.nasa.gov/observatory.html https://wfirst.ipac.caltech.edu/sims/Param_db.html Table 1.
The WFC imaging filters: Central Wavelengths, Width,Zero-points, and average PSF FWHM values.Filter Central Filter AB PSFWavelength FWHM Zero-point a FWHM( µ m) ( µ m) (pixel) Z
087 0.87 0.22 26.39 1.69 Y
106 1.09 0.27 26.41 1.86 J
129 1.30 0.32 26.35 2.12 H
158 1.60 0.40 26.41 2.44 F
184 1.88 0.31 25.96 2.71 W
149 1.40 1.1 27.50 2.19 a Here the zero-point is calculated using each filters effective areaand is equivalent to the magnitude that results in one count persecond for an infinite aperture. (IFC). The WFC possesses an imager and has the abilityto perform slit-less grism spectroscopy, while the IFC hastwo small-field integral field units (IFUs). Combined, theseinstruments will be used to perform the dark energy sur-vey, as well as the Micro-lensing, and High Latitude Surveys.
The Wide Field Channel:
In its most simplistic formthe optical layout of the WFC consists of three mirrors, twofold mirrors, and an eight slot filter wheel. Currently, six ofthese slots are dedicated to imaging filters, one is for a grismthat will provide low-resolution spectra of the full WFC FoV,and the last is a dark filter for calibration. An additional slotto the filter wheel has also been proposed, which will be usedfor either a bluer or redder imaging filter.Eighteen 4k ×
4k HgCdTe detectors (H4RG-10) will beused by the WFC, and will be arranged into a 6 × of 0.281 deg .The six imaging filters of the WFC are named Z Y J H F W µ m respectively, and combinedcover the 0.76 – 2.0 µ m range, as illustrated in Figure 1 .The spatial resolution of the imaging component of theWFC is ∼ (cid:48)(cid:48) pixel − with an inter-pixel capacitance of0.02 in each of the four neighboring pixels. The gain forthe WFC is assumed as unity. A more detailed descrip-tion of the WFC filters including their zero-points andfull-width half maximum (FWHM) can be found in Table 1.The WFC grism is designed such that it providesspectroscopic coverage within the 1.35 – 1.89 µ m range.It possesses a dispersion of 1.04 – 1.14 nm pixel − , with aspectral resolving power of λ/ ∆ λ ≈
622 – 871 (2 pixels).However, we do not focus on the use of the grism in this See https://wfirst.ipac.caltech.edu/sims/Param_db.html?csvfile=WFirstParameters_v5.0.csv for a list of more detailedWFI parameters. More filter information is provided within https://wfirst.gsfc.nasa.gov/science/sdt_public/wps/references/instrument/WFIRST-WFI-Transmission_160720.xlsm - pages 5through 10. As the WFIRST PSF is non-gaussian, the PSF FWHM valuespresented and used for this analysis are derived from the noise-equivalent areas.MNRAS000
622 – 871 (2 pixels).However, we do not focus on the use of the grism in this See https://wfirst.ipac.caltech.edu/sims/Param_db.html?csvfile=WFirstParameters_v5.0.csv for a list of more detailedWFI parameters. More filter information is provided within https://wfirst.gsfc.nasa.gov/science/sdt_public/wps/references/instrument/WFIRST-WFI-Transmission_160720.xlsm - pages 5through 10. As the WFIRST PSF is non-gaussian, the PSF FWHM valuespresented and used for this analysis are derived from the noise-equivalent areas.MNRAS000 , 1–29 (2016)
R. Hounsell et al.
Figure 1.
WFIRST
WFC imaging filter bandpass effective areas, A eff , divided by the maximum effective area (solid lines) as describedby the WFIRST
Cycle 6 instrument parameter release. Also shown are the
HST
WFC3 filters used for this work (dotted lines). TheWFC3 throughputs presented here have been scaled for comparison. paper.
The Integral Field Channel:
The IFC containstwo image slicers that feed a spectrograph. Each imageslicer corresponds to a different FoV: the smaller FoV,higher spatial-resolution IFC-S, which is designed for SNobservations, and the larger FoV, lower spatial resolutionIFC-G, which is designed for galaxy observations (unrelatedto the SN survey). The IFC-S has a 3.00 (cid:48)(cid:48) × (cid:48)(cid:48) FoVthat is composed of 0.15 (cid:48)(cid:48) wide slices, a 0.05 (cid:48)(cid:48) pixel − plate-scale, and a wavelength range of 0.42 – 2.0 µ m. Theinstrument has a spectral resolution of λ/ ∆ λ ≈ . See page 13 of https://wfirst.gsfc.nasa.gov/science/sdt_public/wps/references/instrument/WFIRST-WFI-Transmission_160720.xlsm for more IFC-S in-formation.
Figure 2.
WFIRST
IFC-S throughput for applicable wave-lengths. Note that wavelengths beyond those displayed have nothad their throughputs calculated. MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey The
WFIRST
SDT final report (Spergel et al. 2015)presents a SN survey strategy in which the imaging compo-nent of the WFC is used for SN discovery and the IFC-Sfor classification and obtaining distances. An outline of thisstrategy is described. • The SN survey is a 2-year survey of a single SN field(the field location is still to be decided, and there may be apossibility that two separate fields are selected), with a 5-daycadence. There are therefore 146 visits to the SN field. • Each visit, or epoch of observation, is 30 hours long,meaning the total survey time is 4380 hours (6 months),including overheads. • Within each visit 8 hours of imaging is used exclusivelyfor SN discovery. These data are obtained every 5 observer-frame days. • The imaging is split into 3 sub-surveys (hereafter re-ferred to as tiers) of differing area/depth, and using differentdiscovery filters (see Table 2). • The remaining 22 hours in each visit are for IFC-S ob-servations, used to classify the SN and to synthesise broad-band photometry. • IFC-S observations are designed to be taken at a ca-dence of roughly 5 rest-frame days, with the goal of obtain-ing spectrophotometry to measure distances. • There are 3 different kinds of IFC-S exposures: typicalshort exposures, medium classification exposures, and long“deep” exposures. These 3 exposures represent the first threeIFC-S spectra taken for each SN detected. • The short and medium spectra are used for initial classi-fication, and if these spectra meet certain criteria (outlinedin more detail below) the IFC-S obtains a long exposurethrough which a final classification is obtained. If determinedto be a SN Ia, further followup is initiated • The followup consists of six short spectra plus onemedium exposure of the host-galaxy, taken after the SN hasfaded, to use as a template. • The exposure times for the long and medium spectraare approximately 1.8 and 1.3 times longer than the shortexposure respectively. • The total set of observations for any given SN Ia isequivalent to ∼ z < .
4, over an area of 27.44 deg , using the Y + J filters for discovery. The second is a medium tier forSNe with 0 . (cid:54) z < .
8, over a moderate 8.96 deg area,using the J + H filters. Finally, the last is a deep tier forSNe with z (cid:54) .
7, over a small 5.04 deg area, again usingthe J + H filters. Table 3 lists the exposure times for eachof the three tiers and the number of space-craft pointingsrequired to make up their designated areas. The differentfilter combinations for each tier were chosen in order to probe similar rest-frame wavelengths. However, for theshallow tier the Z -band filter is the only band that coversa rest-frame wavelength range which is sufficiently modeledfor cosmological analysis. One might assume therefore thatredder wavelengths (i.e., > WFIRST orprecursor data.Of the 146 planned visits, the discovery search willbe implemented in only 132. The remaining survey time( ∼
70 days) will be used for host-galaxy follow-up ob-servations i.e., acquiring a template. The host-galaxytemplate spectrum is to be taken a year after the peakbrightness of the SN, when the relative amount of lightfrom the SN compared to the galaxy is negligible. Thus,in the first year only 27 of the total 30 hours in each5-day visit will be used, with the remainder deferred toyear 2. SNe discovered during the second year will havetheir galaxy reference spectrum taken in year 3, after thediscovery component of the 2-year SN survey has concluded.The spectroscopic observations planned in the SDTreport are designed to observe one SN at a time, using theIFC-S. The exposure times were tailored to achieve a signalto noise ratio (SNR) high enough to clearly identify keyspectral features. The longest exposure times are thereforerequired for the highest redshift SNe, i.e., z ≈ . rd spectrum, core-collapse (CC)SNe are eliminated from the sample, see Section 3.1). A listof exposure times for each SN (excluding the host-galaxytemplate) per redshift tier is given in Table 4.Both the SDT report and Spergel et al. (2013, an earlierSDT publication) assumed a combined slew and settle timefor instruments of 42 seconds. The exposure times listed foreach filter within Table 3, and each redshift bin within Ta-ble 4 do not include this overhead, but are the actual timespent on sky. It should be noted that within Spergel et al.(2013) this 42 second overhead was not removed from theirimaging exposures. As a result of this, their associated totaldepths per filter for the SDT report SN survey were deeperthan expected by a maximum of two magnitudes. To reducefuture confusion, we present updated total depths in Ta-ble 2 . Note also that this 42 second slew and settle time isa severe underestimate of the actual value, which now looksto be a factor of two greater (presented at the October 2016Formulation Science Working Group meeting).The total time ( t tot ) listed per imaging tier in Table 3,including overheads, is therefore calculated as t tot (s) = ( t exp + t oh ) × N f × N p , (4)where t exp is the exposure time on the sky in seconds, t oh is the 42 second overhead, N f is the number of filters used Derived from ETC calculations, see https://wfirst.ipac.caltech.edu/sims/ETC.html
MNRAS000
MNRAS000 , 1–29 (2016)
R. Hounsell et al.
Table 2.
Description of the three tier SN survey as outlined in the SDT report.SurveyTier RedshiftRange Area(deg ) DiscoveryFilters Depth perExposure (mag) Total Depth(mag)Shallow 0 . (cid:54) z < . Y, J . (cid:54) z < . J, H . (cid:54) z (cid:54) . J, H
Table 3.
Exposure times ( t exp ) and number of pointings ( N p ) foreach filter within each redshift tier of the survey. The exposuretimes listed here do not include the 42 second slew.Survey Y -band J -band H -band N p t tot Tier t exp (sec) t exp (sec) t exp (sec) (hours)Shallow 13 13 0 98 3.0Medium 0 67 67 32 2.0Deep 0 265 265 18 3.0 Table 4.
Short, medium, and long exposure times ( t exp ) per 0.1redshift bin for the WFIRST
IFC-S component. The far right-hand column lists the total time ( t tot ) spent observing a SNwithin a given 0.1 redshift bin (not including the template host-galaxy spectrum).MeanRedshiftBin Short t exp (s) Medium t exp (s) Long t exp (s) t tot (s)0.15 27.39 36.98 49.30 278.000.25 58.22 78.60 104.80 590.960.35 103.11 139.20 185.60 1046.580.45 170.28 229.88 306.50 1728.320.55 255.44 344.85 459.80 2592.760.65 303.50 409.73 546.30 3080.530.75 354.22 478.20 637.60 3595.360.85 397.72 536.93 715.90 4036.880.95 482.11 650.85 867.80 4893.431.05 605.11 816.90 1089.20 6141.881.15 722.39 975.23 1300.30 7332.251.25 872.11 1177.35 1569.80 8851.931.35 1043.67 1408.95 1878.60 10593.221.45 1200.11 1620.15 2160.20 12181.131.55 1350.33 1822.95 2430.60 13705.881.65 1422.33 1920.15 2560.20 14436.68 (which for discovery is always 2), and N p is the number ofpointings. The detection and selection of SNe Ia for follow-up ob-servations as outlined in the SDT report is a complex pro-cess, influenced by the costliness of single-object follow-upobservations with the IFC-S. The process starts with all pos-sible SNe, both SNe Ia and CC SNe, and then progressivelyremoves SNe which do not satisfy certain conditions. Thefirst part of this selection procedure involves a SNR require-ment. Although it is not clear within the SDT report if thisrequirement is based on image subtracted data, we assumefor this paper that it is. Note also that pre-existing spectro-scopic redshifts for all host-galaxies are assumed by the SDTreport, thus enabling the classification procedure outlined.At each stage of the selection process SNe are removed, andcannot re-enter. Therefore, each step in the selection processis considered a set of selection cuts, which we list below. • Cut 0:
Objects are “detected” if they have a SNR (cid:62) Y + J or J + H ), within asingle epoch (the exact origin of this SNR value is ambigu-ous). SNe which do not satisfy this SNR requirement are notconsidered for follow-up observations. Those SNe which are“detected” are then subject to further constraints. • Cut 1:
Objects that have discovery-epoch colours in-consistent with being a SN Ia at their host-galaxy redshiftare removed. All remaining objects are scheduled for a shortIFC-S spectrum during the next visit to the SN field. • Cut 2:
Objects that do not brighten between the firstand second epochs, or present colours that are consistentwith a SN Ia at the assumed redshift are removed. All re-maining objects are scheduled for a medium IFC-S spec-trum. • Cut 3:
After obtaining the medium spectrum, an ob-ject that does not continue to rise, have consistent colours,nor present a spectrum consistent with that of a SN Ia, is re-moved from the sample. All remaining objects are scheduledfor a long IFC-S spectrum. • Cut 4:
An object that is not confirmed as a SN Ia withthe long IFC-S spectrum is removed. Remaining objects arescheduled for follow-up observations and are included in thefinal cosmology sample.
The survey strategy presented by the SDT report is de-signed such that statistical uncertainties match the assumed“optimistic” systematic uncertainty budget. This means thatthe assumptions about systematic uncertainties set the pa-rameters of the entire project, as they dictate the desiredsample size, which in turn sets the required discovery rateand redshift distribution. The final distribution of SNe Ia per0.1 redshift bin, as expected by the SDT report, is shown inFigure 3 (left panel).The systematic uncertainties presented in the SDT re-port for the
WFIRST
SN survey follow the description ofdistance modulus uncertainties used for the SNAP designoutlined by Kim et al. (2004) (see also Perlmutter & Schmidt2003; Frieman et al. 2003). In the SDT report, the magni-tude of the uncertainties was reduced roughly by a factorof two compared to the SNAP design. The formulation as-sumes that the systematic uncertainties are uncorrelated onscales larger than ∆ z = 0 . σ sys = 0 . z )1 . . (5)However, there are known systematics which contradict thisassumption. Specifically uncertainties related to calibrationand SN colour are correlated across a wide redshift range.The SDT systematic model (Equation 5) is overly simplistic MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey Figure 3.
Left : Redshift distribution of the
WFIRST
SN survey presented (and assumed) by the SDT report.
Right : Fractional statistical(black curve), systematic (red dot-dashed curve), and total (blue dashed line) distance uncertainty per ∆ z = 0 . and not used for our analysis. The SDT functional form forthe systematic uncertainty model also drives the broad, flatredshift distribution seen in Figure 3 (right panel).The SDT report assumes that the distance precisionper SN is σ meas = 0 .
08 mag, and this includes both sta-tistical measurement uncertainties and statistical model un-certainties. This uncertainty is a constant since the SDTstrategy adjusts the exposure time for each SN observa-tion based on redshift so that all SNe have approximatelythe same distance uncertainty. The intrinsic scatter in cor-rected SN Ia distances is set to be σ int = 0 .
09 mag. Thisvalue is more optimistic than what is currently measuredfor optical data where σ int (cid:39) .
13 mag (see Section 7.1 ofKessler & Scolnic 2016). The lensing uncertainty is modeledas σ lens = 0 . × z mag, which is an average of the valuesderived by Holz & Hughes (2005); Gunnarsson et al. (2006);J¨onsson et al. (2010). The total statistical uncertainty for agiven redshift bin is therefore given in the SDT report as σ stat = ( σ + σ + σ ) / N / (mag) , (6)where N SN is the number of SNe Ia in a given redshift bin.The statistical, systematic, and combined uncertaintybudgets of the SDT report SN survey are illustrated in Fig-ure 3 (right panel). To be clear, the SDT analysis is notbased on SN simulations or light-curve analysis, but insteadis based on assumptions about statistical and systematic un-certainties that would arise from such an analysis. Within this paper, we test the various assumptionsmade by the SDT report for the SN survey, evaluate its sta-tistical and systematic error budget, and develop a frame-work to explore other strategies and optimise parameters forthe future
WFIRST mission. To accomplish this, we simulateand analyse a realistic survey and include the most signifi-cant uncertainties. Here we describe software tools that we have used to implement the simulation, apply selection crite-ria, and determine cosmological constraints used to computethe FoM.To examine a variety of possible
WFIRST survey strate-gies, we used the SNANA simulation package (Kessler et al.2009a) . SNANA is a powerful tool that has been extensivelyused for the simulation of SN surveys and analysis of SNsamples (see e.g., Betoule et al. 2014; Scolnic et al. 2014b).The goal of the WFIRST
SNANA simulation is to providethe same fidelity as an ideal image-level simulation by us-ing image properties (zero-points, sky noise, PSFs) ratherthan images themselves. As this is a “catalogue-level” simu-lation rather than a pixel-level simulation, we assume thatPoisson noise correctly describes the uncertainties from theimage-subtraction.To characterise a
WFIRST
SN strategy, we provideSNANA with information about the observatory (e.g.,filter/spectrograph properties and noise sources), the sur-vey (e.g., cadence, exposure time, selection requirements),and the physical Universe (e.g., SN spectral models, SNrates, cosmological parameters, lensing assumptions). Eachof these components is described below in addition toexternal processes which lead to FoM determination. Ouranalysis has resulted in several publicly available upgradesto SNANA.
Imaging filters and spectroscopic bins:
Tables 1 andA1 (in Appendix A) describe the WFC imaging filters andIFC-S wavelength bins used within our simulations. SNANAwas originally designed only to simulate broad-band SNlight curves. In order to simulate the IFC-S, we added anew SNANA module for simulating spectra and “synthetic”broad-band filters.While it may be possible to directly infer distancesfrom SN spectral time series, examination of that approach http://snana.uchicago.edu MNRAS000
Tables 1 andA1 (in Appendix A) describe the WFC imaging filters andIFC-S wavelength bins used within our simulations. SNANAwas originally designed only to simulate broad-band SNlight curves. In order to simulate the IFC-S, we added anew SNANA module for simulating spectra and “synthetic”broad-band filters.While it may be possible to directly infer distancesfrom SN spectral time series, examination of that approach http://snana.uchicago.edu MNRAS000 , 1–29 (2016)
R. Hounsell et al. is beyond the scope of this paper. Instead, we implementthe SDT report’s IFC-S strategy in SNANA by integratingeach simulated spectrum into a set of 52 synthetic filters.These synthetic filters were determined by binning togetherthe 352 spectral elements of the IFC-S by a factor of ∼ Cadence and exposure time:
The cadence of both theWFC and IFC-S components of the SN survey are describedin Section 3. The exposure time per imaging tier of thesurvey is given in Table 3, with IFC-S redshift dependenttimes presented in Table 4. The exposure time of theIFC-S within a given 0.1 redshift bin is identical betweenimaging tiers. Our simulations do not make adjustments toaccount for the mean SN brightness shifting slightly withina redshift bin (i.e., changes in brightness at z = 0 .
45 to z = 0 .
46 etc) as it is unlikely that any actual SN surveyexecuted would have specific exposure times for individualobjects of interest.
Sources of noise:
For all simulated SN observations, weinclude four sources of noise: zodiacal light, thermal back-ground, dark current, and read noise. The contributions fromeach of these sources are presented in Tables 5, 6, and A1,within Appendix A. Host-galaxy Poisson noise is also in-cluded in both the SN-search and template observations,where possible.The zodiacal light is calculated using a broken powerlaw as described in Aldering (2001). Thermal noise contri-butions are calculated using code developed by D. Rubin(private comm.) under the assumption of a 260 K operatingtemperature, and are comparable to values produced whenusing the
WFIRST
ETC . The zodiacal and thermal noisefor the IFC-S, as a function of wavelength, are presented inTable A1 in Appendix A. The higher resolution of the IFC-S leads to smaller zodiacal and thermal noise contributionswhen compared to the WFC.We assume a dark current for the WFC of0.015 e − s − pixel − (Hirata 2014), and for the IFC-S0.003 e − s − pixel − (a conservative estimate based on cur-rent measurements of 0.001 e − s − pixel − ). The read noiseis a function of exposure and read-out time and is calcu-lated using a modified version of the expression described byRauscher et al. (2007). For any given WFC exposure time, See https://wfirst.ipac.caltech.edu/sims/ETC.html
Table 5.
The WFC imaging filters: Sources of noise.Filter Zodiacal Noise Thermal Noise(e − s − pixel − ) (e − s − pixel − )Z087 0.34 0Y106 0.38 0J129 0.36 0H158 0.35 0.005F184 0.20 0.125W149 0.97 0.099 Table 6.
Read noise for each tier of the WFC imaging survey.Calculated via Equation 7Survey Tier Read Noise(e − pixel − )Shallow 26.38Medium 14.53Deep 8.67 the read noise, σ read , is σ read (e / s) = (cid:115)
25 + 4800 × ( t exp /t read − t exp /t read ) × t exp /t read + 1)(7)where t exp is the exposure time of the observation in secondsand t read is the read time in seconds, which is taken as 2.825seconds.For each SN, the underlying sky and host-galaxy fluxis constant in time, meaning that the associated “template”noise for a SN is coherent across exposures. The inclusionof this noise source is particularly important to the analysisof IFC-S observations. In the SDT report each template isplanned to be a single medium exposure. As this exposureis not particularly long (and shorter than the long expo-sures), it adds significant noise to the template-subtractedSN spectrophotometry. On the other hand, this source isnegligible for the WFC photometry, as imaging templatescan be generated from several images, significantly reducingthe template noise.For each WFC simulated SN, we draw an underlyinghost-galaxy flux from a distribution determined from thehigh- z HST
SN survey portion of the CANDELS (Groginet al. 2011; Koekemoer et al. 2011) program. From theCANDELS SN sample, we determine the host-galaxy sur-face brightness at the SN position for a 0.2 (cid:48)(cid:48) radius aperturein the F W , F W , F L , F W , F W , F W ,and F W HST filters. We then fit spectral models to thehost-galaxy measurements. From this sample, we determinethe expected flux in each
WFIRST filter as a function ofredshift. SNANA has the ability to add host-galaxy flux fora variety of galaxy profiles and brightnesses. Since we havemeasured the flux at the SN position, we force the SN posi-tion to be at the center of an appropriate-brightness galaxywith a Sersic profile of index 0.5.SNANA cannot currently simulate host-galaxy Poissonnoise for IFC-S spectra or synthetic filters. However,investigations of this noise in WFC simulations shows thatthis is a negligible ( < Volumetric SN Rates:
To accurately determine the num-
MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey ber of SNe Ia (and CC SNe) that can be discovered by WFIRST , we parameterize the rate as a function of red-shift, and fit to rate measurements that extend to z = 2 . Ia ( z ) = (cid:40) . × (1 + z ) . (10 − yr − Mpc − ) , for z < . . × (1 + z ) . (10 − yr − Mpc − ) , for 1 < z < . (8)Similarly, we use the Strolger et al. (2015) CC SN rate,R CC ( z ) = 7 . × (1 + z ) (10 − yr − Mpc − ) . (9)As the expected detection rate for z > Spectral models:
We base all of our SN Ia simulationson the SALT2 spectral model Guy et al. (2010). Accuratespectrophotometry can be produced from this model, cov-ering a range of phases and light-curve shapes. The SALT2model is parameterised by x , its light-curve shape param-eter. This parameter adjusts the brightness as a function ofwavelength and phase, simultaneously changing the light-curve and spectral shape. The spectrum is further adjustedby a colour law, which coherently changes the spectral shapeat all epochs, with the amount of colour change being param-eterised by the SALT2 parameter c (where the rest-frame B − V colour is highly correlated with c ).One can determine the distance to a SN Ia with mea-surements of x , c , and the log of the fitted SN Ia amplitude, m B , through a Tripp (1998) formulation, µ = m B − M + α · x − β · c, (10)where µ is the distance modulus, α and β are hyperparam-eters (generally fit to minimize the Hubble residuals for anentire sample) that dictate the relation between absolutemagnitude and x and c , respectively, and M is the abso-lute B -band magnitude of a fiducial SN Ia with x = 0 and c = 0. Therefore, x and c determine the apparent bright-ness for a particular SN Ia given its redshift, cosmologicalparameters, α , β , and M .The specific SN Ia spectral model used in our simula-tions to generate and fit our SN light curves and spectra, isan extension of the SALT2 SN model (Guy et al. 2007, 2010).While WFIRST will observe SNe in the rest-frame NIR, thefiducial SALT2 model is limited to optical wavelengths. Toextend the model to the NIR, we follow the same procedureas was used for the simulations of SNe for the CANDELSand the Cluster Lensing And Supernova survey with Hubble(CLASH) (Rodney et al. 2014; Graur et al. 2014; Strolgeret al. 2015).This SALT2 extrapolation uses a compilation of 118,well-sampled, low- z SNe Ia with both optical and NIR lightcurves (Avelino et al. in prep.; Friedman et al. in prep.).NIR light curve data are obtained from nearby SN surveys,principally from CfA IR1-2 (Wood-Vasey et al. 2008; Fried-man et al. 2015), and CSP (Contreras et al. 2010; Stritzingeret al. 2011), as well as other sources (see Table 3 of Friedmanet al. 2015, and references therein). Corresponding opticalphotometry comes largely from CfA1– 4 (Riess et al. 1999;Jha et al. 2006; Hicken et al. 2009a, 2012), CSP (Contr-eras et al. 2010; Stritzinger et al. 2011), and LOSS (Gane-shalingam et al. 2010). Each SN light curve in this sample is used to generate a spectrophotometric model by warping theSN Ia spectral template from Hsiao et al. (2007) to matchthe observed photometric colours at each epoch.From the resulting set of 118 warped spectral time se-ries models, a median spectral-energy distribution (SED) isderived for each phase, and smoothly joined with the 0 th -order component of the SALT2 model (the M componentin Guy et al. 2007). The higher order SALT2 model com-ponents, including variance and covariance terms, are ex-trapolated using flat-line extensions . This model has notyet been calibrated to produce accurate distance estimatesfrom real data. However, this SALT2 extrapolation is suffi-cient for producing realistic simulations for the purposes ofinvestigating the WFIRST
SN survey optimisation.Finally, we extrapolated the SALT2 colour law to in-frared wavelengths using a modification of the polynomialfunction from Guy et al. (2010). The polynomial coeffi-cients were set so that the effective colour law approximatelymatches the extinction curve of Cardelli et al. (1989), with R V = 3 . ∼
70% lumi-nosity variation and 30% colour variation. This model intro-duces 0.13 mag of scatter to the Hubble diagram. While theSDT report assumes the intrinsic scatter is entirely achro-matic, the scatter model used here does not. The populationparameters for the colour and stretch distributions of oursimulations are those derived in Scolnic & Kessler (2016)for the high- z SN sample.The CC spectral models used within our simulationsare described in Kessler & Scolnic (2016); Kessler et al.(2010), and were generated from a combination of SDSS(Sako et al. 2014b) and CSP (Hamuy et al. 2006) light-curvedata. The NIR CC templates are extrapolated from theNugent CC template spectra . Jones et al. (2016) analysedthe effect of varying the assumed CC SN luminosityfunction (Li et al. 2011) on the observed distribution (e.g.,discovery rate as a function of redshift), and found thatthere are differences between the simulated observables anddata. However, they also found that varying the luminositydistribution had an insignificant effect on the final SN Iapurity of the photometric sample Selection requirements (cuts):
Within the SDT report, aSN (both Ia and CC) is detected if it has an observation withSNR (cid:62) Y + J or J + H ),within the same epoch. As a precursor to this criteria wesimulate a trigger that requires a SNR (cid:62) J + H or Y + J )light curves of the SNe are analysed via a code outside ofSNANA. This code applies the photometric selection crite-ria defined in Section 3.1. Spectra of the objects that suc-cessfully pass these criteria are analyzed via a modified,NIR-enabled version of the Supernova Identification (SNID;Blondin & Tonry 2007) package. SNID compares each inputSN spectrum to a library of template spectra and determines For more details, see http://github.com/srodney/wfirst See https://c3.lbl.gov/nugent/nugent_templates.html
MNRAS000
MNRAS000 , 1–29 (2016) R. Hounsell et al. how closely template spectra match the input. Within ourwork a SN is typed to be a “good” Ia if 80% of the matchesand the top match are a SN Ia at the correct redshift, andif the SN is discovered roughly 7-12 days before peak. ThisSNID spectral analysis is used to implement the spectro-scopic cuts described in Section 3.1.For imaging-only strategies the selection criteria forinclusion in the final sample occur only in the final analysisi.e., no choices are made during the survey itself. First,we require that each SN have at least one epoch with aSNR (cid:62)
10 and at least two epochs with a SNR (cid:62)
5. Theserequirements are the same as those used to forecast theanalysis of the DES-SN survey (Bernstein et al. 2012). Wenote that these requirements are conservative and accuratedistances can be obtained with less stringent requirements.Next, we require that the light-curve parameters of each SNto fall within a “typical” range of colour and stretch valuessuch as those defined by Betoule et al. (2014, and referencestherein), i.e., − < x < − . < c < . Lensing:
We set the distance uncertainty caused by line-of-sight gravitational lensing to match that used by the JointLight-curve Analysis (JLA) and PS1 analyses as describedby J¨onsson et al. (2010), σ lens = 0 . × z (mag) . (11) Extinction:
Since the SN fields have not been cho-sen, we assume that the field will have a low value of E ( B − V ) = 0 .
015 mag (Schlafly & Finkbeiner 2011). Sys-tematic errors associated with this extinction are discussedin Section 8.
Cosmological Priors:
After the simulation, selection cutsare applied and each light curve is fit with the SALT2 model.Typical broadband
WFIRST light-curve fits are shown inFigure 4 for a range of redshifts. The ensemble of fittedparameters ( c , x , m B ) are then used to determine cosmolog-ical constraints through the use of CosmoMC (Lewis 2013).FoMs are calculated corresponding to the inverse area ofthe 95% confidence contours in the w − w a space (Albrechtet al. 2006). For each FoM determination, we assume a flatUniverse and marginalise over H and Ω M . Furthermore,we include constraints from both baryon acoustic oscillation(BAO; Anderson et al. 2014) and cosmic microwave back-ground (CMB; Planck Collaboration et al. 2016) datasets.To provide an anchor to the SN distance measurements,we also include 800 simulated SNe Ia with z < . WFIRST , which we model as hav-ing the characteristics of the Foundation SN survey (Foleyet al., in prep). The Foundation survey uses the PS1 tele-scope and observes low- z (0 . < z < .
1) SNe in griz every5 days with typical distance uncertainties < . z SN Ia sample is a requirement specifiedin the SDT report.When changing the assumptions for systematic uncer-tainties we use a modified version of CosmoMC to reducethe computational complexity of determining the FoM. Thisversion of CosmoMC, which we call “CosmoMC*”, encodesthe CMB information using the compressed Gaussian likeli-hood presented by the Planck Collaboration et al. (2016, seetheir Table 4; the version which does not marginalise over A L ) and only accounts for the geometric effects of dark en- ergy. Therefore CosmoMC* fixes τ , the re-ionisation opticaldepth, and log( A ) (equivalent to ln(10 A s ), where A s is theinflation power spectrum amplitude). These changes signif-icantly reduce the time to compute the FoM. We have doneextensive checks to ensure that the CosmoMC* model pro-duces accurate results relative to the original version, thatuses the full set of Planck likelihoods. Fluctuations of a fewpercent in the FoM value is expected, reflective of the levelof convergence for the MCMC chains.Using all aforementioned tools, we have performed thefirst set of realistic simulations and analysis for the WFIRST
SN survey. Through examination of the data produced wehave been able to determine statistical and systematic un-certainties, with a variety of data-driven choices.
Here we describe simulations of several different strate-gies for the
WFIRST
SN survey. The survey variations ex-amined here are summarised in Table 8. This table lists thevarious strategy names, filters, imaging tiers, areas, and theresultant number of simulated SN Ia analysed.We analyse strategies that use both the WFC imagerand the IFC-S spectrograph (the SDT, SDT* and SDT*Highz strategies) as well as strategies that employ imagingexclusively (the Imaging, Imaging:Lowz, and Imaging:Highzstrategies).For each variant on the
WFIRST
SN strategy, we re-main constrained by the 6 months total observing time. Fur-thermore, for the imaging component of the survey, the ex-posure time per tier, filter zero-points, and sources of noisefor each filter are specified in Tables 1, 3, 5, and 6. Whenthe IFC-S is used, its bandpass, redshift dependent exposuretimes, and sources of noise remain set to the values given inTables A1 and 4.If an instrument (i.e., the IFC-S), tier (shallow or deep),or filter within a survey simulation is removed or added, theareas (listed in Table 2) of the remaining tiers are adjustedevenly (except in the SDT* Highz case; see Section 5.2) toaccount for the loss or gain of time.Note that we have not changed the cadence, depth ofa given tier, IFC-S strategy (epochs and number of SNe),or
WFIRST filter bandpass for any strategy outlined withinthis paper. Such investigations/optimisations will be the fo-cus of future papers.For strategies that only have an imaging component,we consider the impact that additional non-
WFIRST fil-ters would have on the survey. For simplicity, we assumethat the additional filters are similar to those from
HST’s
WFC3, and as such we have used their throughputs andtaken the average AB magnitudes of the two WFC3 chips tobe our zero-points (see Table 7). The FWHM values for thesefilters are calculated in part via the use of the WebbPSFfor
WFIRST tool. This tool allows the user to input ap-propriate SNe spectra, and account for wave-front aberra-tions, in order to calculate binned and un-binned PSF data.WebbPSF however, is not designed for filters bluer than the Z -band. We therefore modified this tool to calculate bluer https://pythonhosted.org/webbpsf/wfirst.htmlMNRAS , 1–29 (2016) imulations of the WFIRST SN Survey z = 0.5 Z Y J F l u x H T obs − T F z = 0.9 Z Y J F l u x H T obs − T F z = 1.3 Z Y J F l u x H T obs − T F z = 1.7 Z Y J F l u x H T obs − T F Figure 4.
Example
WFIRST broadband (
ZY JHF ) simulated light curves (black circles) and best-fit light-curves (smooth curve) forSNe at redshifts 0.5, 0.9, 1.3, and 1.7. Magnitudes are 27 . − . × log (Flux); e.g., m = 25 . Table 7.
HST
WFC3 filters included within
WFIRST
SN simu-lations: Zero-points and average PSF FWHM values are listedFilter Zero-Point (AB) PSF FWHM (pixel) F W ( B ) 24.75 1.62 F W ( V ) 25.72 1.62 F W ( R ) 25.44 1.63 F W ( I ) 25.03 1.67 wave-front aberrations via the extrapolation and applica-tion of higher order Zernike coefficients. Pixelation is thenapplied to these results along with an inter-pixel capacitanceeffect on the order of ∼ WFIRST ” defined filters, or our bluer WFC3 like filters.We have not yet investigated the effect of adding filters red-der than the F -band.In the current SDT strategy, a set number of SNe inthe 0 . < z (cid:54) . > WFIRST survey is completed, sincethey are not needed to define the SN follow-up observationalsequence (as is the case for the SDT survey).The design of each survey strategy is discussed below.
Here we present the simulated SDT survey strategy (seeSection 3). We also present a slight modification to the SDTstrategy to significantly improve efficiency (the SDT* strat-egy). These strategies use both WFI channels: the WFCimager and IFC-S.The number of generated SNe is set by the volumetricrates, survey area, depth, and duration; they are reported inTable 9 and do not include selection requirements. Withinthe appropriate redshift ranges a total of 25,214 SNe are gen-erated, 4268 of which are SNe Ia, with the remaining 20,946being CC SNe. The initial SDT SNR requirement describedin Section 3.1, reduces the total to 7,951 “detectable” events(4116 of which are SNe Ia, see Table 10). For these detectableevents, 3,106 pass all of the photometric cuts specified withinthe SDT report (listed in Section 3.1). A breakdown of thenumber of SNe to pass each cut is given in Table 10.Each medium- and long-exposure SN spectrum (whichare obtained for SNe that pass the photometric parts of
Cuts2 and 3, respectively) is compared to a library of real SNspectra using SNID (Blondin & Tonry 2007). The numberof SNe passing these additional spectroscopic selection cri-teria (see Section 3.1) are reported in Table 11. As is donewith all current cosmological analyses, we apply additionalcolour, light-curve shape constraints, which further reducesthe number of SNe in the cosmological sample. After apply-ing all criteria, the final sample consists of 2,214 of the 4,116detected SNe Ia, resulting in an efficiency of only 54%.After applying the initial photometric selection criteria(
Cuts 0 , 1, & 2), the sample has a SN Ia purity of ∼ ∼
22% CC SN contaminants, ∼
59% are SNe Ib/c and ∼
41% are SNe II. The SNe Ib/c that make up the major-ity of the contaminants are also the objects that are mostspectroscopically similar to SNe Ia, and therefore the mostdifficult to remove with low-SNR spectra. Example spectraof a SN Ia that passes all cuts, a SN Ia that is excluded basedon its long-exposure spectrum, and a CC SN (SN Ic) thatpasses all cuts and is included in the cosmology sample are
MNRAS000
MNRAS000 , 1–29 (2016) R. Hounsell et al.
Table 8.
Simulated strategies investigated for the
WFIRST
SN survey. This includes the strategy suggested within the SDT report.
Name Redshift Range Filter Set Used Area (deg2) Number of SN Ia SelectedShallow Medium Deep Shallow Medium Deep Shallow Medium Deep Shallow Medium DeepSDT 0.10–0.39 0.40–0.79 0.80–1.70 IFC-S,
Y J
IFC-S, JH IFC-S, JH Y J
IFC-S, JH IFC-S, JH · · · · · · IFC-S, JH IFC-S, JH · · · · · · Y J JH JH
RZY J RZY J Y JHF · · ·
Y J JH · · · · · · · · ·
Imaging:Lowz* 0.01–2.99 0.01–2.99 · · ·
RZY J RZY J · · · · · ·
822 8117 · · ·
Imaging:Lowz+ 0.01–2.99 0.01–2.99 · · ·
RZY JHF RZY JHF · · · · · ·
588 5167 · · ·
Imaging:Lowz-Blue 0.01–2.99 0.01–2.99 · · ·
BV RIY J BV RIY J · · · · · ·
347 4894 · · ·
Imaging:Highz* · · · · · ·
RZY J Y JHF · · · · · · · · · · · ·
RZY JHF RZY JHF · · · · · ·
Table 9.
Number of SNe generated per SDT survey tier.Survey Redshift Number of Number of Total SNeTier Range SNe Ia CC SNe per TierShallow 0 . (cid:54) z < . . (cid:54) z < . . (cid:54) z (cid:54) . SN Total: 4268 20,946 25,214Table 10.
Number of SNe that make it past the photometric cutsdefined within the SDT report b .Cut Shallow Medium Deep TotalIa CC Ia CC Ia CC0 465 97 1102 428 2549 3310 79511 169 16 867 97 2408 1217 47762 14 1 378 44 2046 623 3106 b Photometric cuts are defined in Section 3.1.
Table 11.
Number of SNe that pass the photometric and spectro-scopic cuts defined by the SDT report, and the number of SNe Iathat pass the required colour and stretch constraints, and as suchare used within our final analysis.Cut Number of SNe Number of SN Iaanalysed3 2445 22564 2316 2214 illustrated in Figure 5. This figure demonstrates the diffi-culty of classification using the SDT report strategy. For thegiven exposure times, distinguishing spectral features cannot be identified with the expected SNR and resolution. Inparticular, the sulfur “W”, which the SDT report uses as aclear example of a SN Ia feature, is not detected in the ex-ample SN Ia spectra of Figure 5 that fails to make it to thefinal sample.The current number of correct spectral classificationsfor the SDT strategy is likely optimistic. While correlatedtemplate noise is included in the simulations, residual noisefrom un-subtracted host-galaxy light as a consequence oflack of a galaxy template at the time of classification, has notbeen included (no template measurements have been madeat the time of classification). Even if a spectrum of the host-galaxy does exist (e.g., from a ground-based spectrograph),the exact galaxy SED at the position of the SN will not beaccurately measured. A galaxy SED could be built from thephotometry or interpolated from surrounding IFC-S pixels,however this will likely introduce significant uncertainties.This source of noise will be explored in future work (seeSection 8).The SN Ia efficiency for the final sample is shown in Figure 6. The efficiency is low at particular redshifts, 0 . 8. This is partially the resultof the survey design producing insufficient SN discoveriesat the the high- z end of each tier. However, photometricselection criteria that require SNe to have colours consistentwith a SN Ia at their host-galaxy redshift, and that theSNe rise between epochs, are the main contributors for thelow efficiency. For the shallow imaging tier, which covers0 . (cid:54) z < . 4, these criteria are problematic due to thetier’s short 13-second exposure, which results in noisy lightcurves. Noisy light curves often do not have a detectable riseat early epochs (i.e., the measured flux in the second epochis often lower than that of the first epoch).The large reduction in the number of SNe between thefirst two spectral epochs is also because of this required in-crease in brightness of a SN between epochs ( Cut 2 ). In manycases statistical noise causes a SN to appear to fade betweentwo successive epochs. To reduce this bias, this criterion isloosened via the iterative examination of a range of“rise”val-ues (including negative values) for the simulated SNe Ia as afunction of redshift for each tier, and applied within our finalresults. The constraints on the discovery filter colours ( Y + J for shallow, J + H for medium and deep) are also tightened,excluding some of the most extreme SNe Ia from the finalsample, and significantly reducing the number of CC SNeat each step. The effect of these improved selection criteriaenables a reduction in the fraction of SNe Ia missed to ∼ z . The SDT* strat-egy, which allows for the possibility of a measured declinebetween early epochs, does not have this problem. Spectro-scopic classification for the SDT* strategy, however, suffersfrom the same issues as the SDT strategy, reducing the effi-ciency to ∼ ∼ 99% pu-rity, which will further increase when considering full lightcurves and all spectral data.To both accurately match the SDT description of their MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey R e l a ti v e f λ Ia Passed Short Medium(Classification) S II "W" Si II λ Long(Peak) Ia Did Not Pass SN Ia Template Non − SN Ia Template S II "W" Si II λ Rest Wavelength (Å) CC That Passed Figure 5. Simulated rest-frame WFIRST IFC-S spectra of z = 1 SNe. The left panels correspond to a SN Ia that passes all cuts andfor which full follow-up observations would be obtained. The middle panels correspond to a SN Ia that is not identified as a SN Ia basedon its long-exposure spectrum and is thus culled from the sample. The right panels correspond to a CC SN that passes all requirements,including photometric cuts, and would receive full follow-up observations. The top, middle, and bottom rows correspond to short-,medium-, and long-exposure spectra, respectively, for each SN. The WFIRST spectra are plotted as blue points with error bars. Notethe changing resolution with wavelength. The best-matching SN Ia and non-SN Ia spectra are plotted as gold and red, respectively. Figure 6. SDT and SDT* (left and right panels, respectively) SN Ia selection efficiency as a function of redshift. The gold squares,green diamonds, blue triangles, and black circles represent the efficiency of SNe Ia that could be scheduled for 1, 2, 3, and 9 epochs ofspectroscopy, respectively. Lines connect data from the same tier of the survey. The large drop in efficiency from the 1 st to 2 nd spectralepoch at z < .000 SDT and SDT* (left and right panels, respectively) SN Ia selection efficiency as a function of redshift. The gold squares,green diamonds, blue triangles, and black circles represent the efficiency of SNe Ia that could be scheduled for 1, 2, 3, and 9 epochs ofspectroscopy, respectively. Lines connect data from the same tier of the survey. The large drop in efficiency from the 1 st to 2 nd spectralepoch at z < .000 , 1–29 (2016) R. Hounsell et al. survey strategy and to reduce potential biases of selectingSNe at the same redshift with different filters (and SNR),we only select SNe at particular redshift from their corre-sponding tiers. This is particularly important for z < . Y + J filters, whiledeeper tiers use the J + H filters. To be specific, SNe with z < . 4, 0 . (cid:54) z < . 8, and 0 . (cid:54) z (cid:54) . z < . z = 0 . z < . z SNe Ia in the final SDT* sample. As the shallow tier did not yield many SNe for the sim-ulated SDT* survey (Section 5.1), we examined the effectsof removing that component and reallocating the time tothe medium tier. This IFC+imaging based strategy there-fore consists of only two tiers: medium and deep. This sim-ulation is designed such that the medium tier is allowed tosample SNe within a greater redshift range, 0 . (cid:54) z < . . (cid:54) z < . ∼ This simulation is based on a worst-case scenario wherethe SDT strategy is executed, but after the fact it is de-termined that the IFC-S data is unusable, resulting in ex-clusive use of the existing WFC imaging data. Presumablythis analysis can happen even if the IFC-S works perfectly.There is no increase in the areas of this strategy as it isexploring the idea of data obtained when an instrument is“faulty”. There are also no SN selection criteria as outlinedin Section 3.1 as there are no spectra. Purity of the resultingSN Ia sample is implemented via the aforementioned SNRrequirements made on fitting (see Section 4). The results ofthis simulated survey are presented in Figure 7c and 7h.The number of SNe Ia obtained within the simulation isa factor of ∼ z < . z SNe in the shallow tier of the survey. This simulated WFC imaging-only survey uses all threeSDT tiers, but four broad-band filters instead of two. Thefour filters used are ZY J and F W ( R -band) for the shal-low and medium tiers, and Y JHF for the deep. For each tierthe area has been adjusted (factors of ∼ ∼ . (cid:54) z (cid:54) . 7, while the Imaging:Allz data set wouldcover 0 . (cid:54) z (cid:54) . 0. Within the same redshift range theSDT* residuals are on average a factor of two greater. Atthe high- z end however ( z > . ∼ × grater than the av-erage SDT* residual. Neither sample has been corrected fordistance bias (as shown in Kessler & Scolnic 2016), whichleads to some of the larger residual values as seen in theSDT* data at low- z . A simulated WFC imaging survey that consists of theshallow and medium tiers only, and the use of the two SDTdiscovery filters ( Y + J and J + H ). The area of the shallowtier is increased by a factor of ∼ ∼ ∼ MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey Figure 7. Left panels : Redshift distributions for each simulated WFIRST SN survey examined. For comparison, the SDT requiredredshift distribution (as assumed by the SDT report) is presented as a grey histogram in the top-left panel (this is equivalent to thatdisplayed in Figure 3). In the same panel, we present the “possible” SDT redshift distribution, which corresponds to all simulated SNe Iadiscovered that pass all SDT selection criteria, as red circles. The red histogram represents the “actual” SDT redshift distribution, whichcorresponds to the lesser of the maximum number of SNe that can possibly be observed at that redshift and the desired number ofSNe Ia for that redshift bin. A similar curve and histogram for the SDT* strategy are shown as blue triangles and a blue histogram.The remaining left panels present the redshift distributions of the other strategies examined. For comparison, the SDT* possible curveis presented as blue triangles in each panel. Right panels : Fractional statistical distance uncertainties for each simulated WFIRST SNsurvey as a function of redshift. For comparison, the assumed SDT uncertainties are plotted as the thick black line in the top-right panel(see Figure 3), with the measured uncertainties from the simulations for the SDT (SDT*) strategies represented by red circles (bluetriangles). The remaining left panels present the fractional statistical distance uncertainties of the other strategies examined, with theleft and right panels of a given row corresponding to the same strategies. For comparison, the “actual” SDT* distance uncertainties arepresented as blue triangles in each panel.MNRAS000 SNsurvey as a function of redshift. For comparison, the assumed SDT uncertainties are plotted as the thick black line in the top-right panel(see Figure 3), with the measured uncertainties from the simulations for the SDT (SDT*) strategies represented by red circles (bluetriangles). The remaining left panels present the fractional statistical distance uncertainties of the other strategies examined, with theleft and right panels of a given row corresponding to the same strategies. For comparison, the “actual” SDT* distance uncertainties arepresented as blue triangles in each panel.MNRAS000 , 1–29 (2016) R. Hounsell et al. Figure 8. Hubble diagram of the simulated WFIRST SDT* sam-ple (black points), and the simulated WFIRST Imaging:Allz sam-ple (blue points). The red and gold lines represent ΛCDM and w CDM (with w = − . 05) models, respectively. The bottom paneldisplays the Hubble residuals relative to the ΛCDM model. Theblack triangles, and blue squares represent binned residuals forthe SDT*, and the Imaging:Allz data, respectively. the number of SNe Ia compared to the final possible samplein the SDT* scenario (see Figure 7d and 7i). Below z < . ∼ z < . 6. The fraction of SN Ia with z > . ∼ Here is another imaging-only simulation that consistsof the shallow and medium tiers, where the area of eachrespective tier has been increased by a factor of ∼ ∼ RZY J filters are used in both tiers, maximisingour coverage of the rest-frame optical and extending to therest-frame NIR. On comparison to the Imaging:Lowz simu-lation we have included an additional two filters, R + Z , andthus decreased the observed areas. For this Imaging:Lowz*simulation the redistribution of IFC-S time, bluer filter, lackof the IFC-S component and selection criteria, resulted in a ∼ ∼ z (cid:62) . Six filters, RZY JHF , are used within this imaging-only, two tier simulated survey (each tier uses all six filters).The area of the shallow and medium tiers are increased byfactors ∼ ∼ ∼ (cid:62) . This simulation is the same as Imaging:Lowz+, howeverbluer filters have been selected, which include the WFC3 F W ( B ), F W ( V ), F W ( R ), and F W ( I ) fil-ters in combination with the WFIRST SDT discovery fil-ters, Y + J . The areas of the shallow and medium tiers areincreased by factors of ∼ ∼ ∼ 5% of the SN sample has z (cid:62) . This simulation is similar to the Imaging:Lowz* strat-egy, however here the medium and deep tiers are used ratherthan the shallow and medium. Time from the IFC-S andshallow components are used to increase tier areas by factorsof ∼ RZY J , with Y JHF for the deep. The number ofSNe Ia found by this strategy is ∼ This simulation is similar to the Imaging:Lowz+ strat-egy, however here the medium and deep tiers are used ratherthan the shallow and medium. The areas of the two tiershave been increased by factors ∼ ∼ H + F in the medium and R + Z in the deep), and thus comparativereduction in area size. See Figure 7e and 7j for the resultsof this strategy. Redshift distributions of SNe Ia and their associateddistance uncertainties are shown in Figure 7. For the Imag-ing:Highz and Imaging:Allz scenarios the number of SN Iadetected per 0.1 redshift bin increases significantly over theSDT* results. However, in the case of each Lowz survey thenumber of SNe Ia detected at z (cid:62) . MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey cuts. The redder filters of this deep tier are also required forthe detection of SNe at higher z .The SDT Imaging and Imaging:Lowz strategies clearlyindicate how ineffective the shallow tier of the SDT SN sur-vey design is. The dearth of SNe Ia at z < . z > 2, but note that ourSN Ia rates are less certain for z > 2. The increased numberof SNe Ia for particular redshift ranges leads to an increase inthe statistical precision per redshift bin, as much as ∼ × better than the SDT* data for the Imaging:Highz* strategy. In addition to the statistical uncertainties examinedabove, several sources of systematic uncertainty have beeninvestigated as part of this analysis. These investigations arethe first attempt to quantify the systematic uncertainties ofthe WFIRST SN survey without the use of ad hoc functionssuch as Equation 5.When considering systematic and statistical uncertain-ties, we compute a covariance matrix to describe the dis-tance uncertainties such that C = D stat + C sys (Conley et al.2011). The first term, D stat , is the purely diagonal matrix,where the diagonal elements correspond to the individual SNdistance uncertainties given by Equation 6. The systematiccomponent, C sys , can be described as the summation overeach systematic uncertainty such that C sys ,ij = (cid:88) k (cid:18) δµ i δS k (cid:19) (cid:18) δµ j δS k (cid:19) (∆ S k ) , (12)where δµ i /δS k is the change in distance modulus for the i th SN when varying the k th systematic effect by ∆ S k . Tocalculate C sys , we determine the distance modulus differencewhen changing each systematic effect by 1 σ . During thisprocess, we fix α and β from Equation 10 to the values foundby minimizing the Hubble residuals when including only thestatistical uncertainties.To understand the dependence of the FoM on the valueof each systematic uncertainty, we introduce a bias in ourmeasurements that mimics the effect of each systematic un-certainty. We vary each systematic effect with multiplicativescaling from 0 (no effect) to 12 times the value of our currentconstraints for that uncertainty. For each case, we comparethe distance moduli determined with the included uncer-tainty to that determined without the effect. We displaythe absolute median distance modulus bias as a function ofredshift for the nominal case (multiplicative factor of 1) inFigures 9 and 10. The µ -differences are used to computethe derivative term in Equation 12. Note that although ab-solute values are presented in Figures 9 and 10, the signsof the differences are unchanged in the computation of thederivative.To determine a FoM, we input the derived distancesand the associated covariance matrix to CosmoMC*. Ad-ditional constrains from CMB and BAO measurements areincluded in the fitting (as discussed in Section 4). The FoM measurement when including a particular systematic uncer-tainty (with various multiplicative scalings), FoM tot relativeto the statistical-only FoM, FoM stat , are shown in Figure 9and 10. The points marked as “current” represent the FoMcalculated with our present understanding of the systematicuncertainty (i.e., a multiplicative scaling of 1), FoM tot , curr .Points marked “optimistic” represent the FoM values calcu-lated with assumptions for improved systematic uncertain-ties, FoM tot , opt . These optimistic systematic uncertaintiesare values which we hope will be available at launch, andhave assumed through reasonable prediction.The limited precision of the CosmoMC* runs (discussedin Section 4) and artifacts of light-curve fitting in SNANAadd some numerical noise to individual FoM measurements,making their values deviate, on order of a few percent, froma smooth interpolation. However, all of our main findingsare robust against these small variations. Calibration uncertainty is currently the largest system-atic uncertainty of all recent ground-based SN cosmologyanalyses (e.g., Scolnic et al. 2014b). The primary sources of calibration uncertainty can be split into three separate com-ponents, which are listed and discussed below. The nominalsize of each component is set to match the current valuesdetermined for the HST system. This is likely a conser-vative assumption, and is varied within the present analysis. The absolute calibration of the spectrophotometricsystem: The accuracy of the HST Calspec system (Bohlinet al. 2014) is described as a linear function with a slope ofroughly 5 mmag per 7000 ˚A (Bohlin 2007). Assuming thefunctional form of the calibration of WFIRST is similar tothat of HST , we use the magnitude of the HST systematicuncertainty as the nominal uncertainty for the WFC (seeFigure 9a and g). For the IFC-S we take it to be 50 mmagper 7000 ˚A, as there are many unknown calibration issues forthis instrument, and as such an estimate ten times greaterthan the WFC is deemed appropriate (D. Law, private com-munication). This higher value for the IFC-S is also appro-priate given work conducted in Bacon et al. (Section 4.6of 2015), which compares synthesized broad-band magni-tudes from MUSE (a panoramic integral field spectrograph)to that of HST , and finds a mean bias of 50 mmag, witha statistical uncertainty of 40 mmag. In addition, similarvalues were also found by Childress et al. (2016) in whichdata from a 2/3 yr SN survey using the Wide Field Spec-trograph on the Australian National University Telescope,enabled the determination of a colour variation ranging from40 mmag in the red to 90 mmag in the blue.For both the IFC-S and WFC we assume an optimistic colour-gradient uncertainty of 3 mmag per 7000 ˚A, ∼ × better than HST . This is the main calibration systematicuncertainty for the IFC-S, and also an uncertainty for thefilters. Non-linearity of the detector: Detector response non-linearity can severely impact photometric precision in as-tronomical observations. Recent work (e.g., Riess 2010) hassuggested that a count-rate dependent non-linearity is com-mon in HgCdTe detectors. WFIRST will be using H4RG de- MNRAS000 MNRAS000 , 1–29 (2016) R. Hounsell et al. Figure 9. Left panels : Median distance modulus residuals for the Imaging:Allz simulated sample of SNe Ia. The residuals are thedifference between distance moduli measured with and without a particular systematic uncertainty applied (at its current value) andscaled to have zero residual at z = 0. Right panels : Fractional FoM values relative to the statistical FoM for different values of a particularsystematic uncertainty (with the scaling being relative to the current value). The dashed line represents the statistical FoM. Red circlesand green squares represent the current and optimistic values of each systematic uncertainty, respectively MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey Figure 10. Same as Figure 9, but for additional systematic uncertainties. tectors for both the WFC and IFC-S. Count-rate dependent non-linearity and its effect in HgCdTe detectors must there- fore be taken into account as a systematic uncertainty forthe WFIRST mission. Riess (2010) measured a non-linearity MNRAS000 MNRAS000 , 1–29 (2016) R. Hounsell et al. in WFC3-IR data of ∼ 1% per dex over a range of 10 mag(4 dex), which was independent of wavelength. We take thisas our baseline assumption (see Figure 9b and h).Our optimistic non-linearity systematic uncertainty isassumed to be 5 × better than the values obtained from HST studies. This is reasonable given current and futureimprovements of detectors. Zero-point uncertainty: Recent ground-based imag-ing surveys have been able to measure the uncertainties intheir filter zero-points to 5 mmag (Betoule et al. 2013; Scol-nic et al. 2015). Because of the colour term in the SN dis-tance equation, a bias in a zero-point can result in a dis-tance modulus bias ∼ × (a multiplicative value equivalentto β ) larger. A space-based observatory, being above the at-mosphere, should have zero-point uncertainties that are atworst equal to the state-of-the-art ground-based surveys. Wehave therefore analyzed this uncertainty via the addition ofa 5 mmag shift to each filter zero-point (see Figure 10). Op-timistic WFC imaging zero-point uncertainties are assumedas a 1 mmag offset. For surveys that use the IFC-S we assume that thereare no CC SNe in the cosmological sample; i.e., we set thecurrent and optimistic contamination to be 0%. Althoughsome CC contamination was present within the SDT/SDT*results (see Section 5.1), we expect that by using all spectraand light curve data available (not just the first 5 imagingdata points and 3 IFC-S spectra), contamination will dropsignificantly. In addition we expect that the application ofimproved classification techniques on these data will furtherimprove classification.For each of our imaging-only scenarios (see Sections 5.3through 5.10) however, contamination on the final cosmol-ogy sample must be considered. When simulating each sur-vey both CC SNe and SN Ia are generated, selection andlight-curve quality cuts are then applied as discussed in Sec-tion 4. Within our work, this results in a photometric clas-sification purity of ∼ 93% (for the Imaging:Allz survey). Toaccount for the systematic uncertainty introduced by any re-maining contamination , Hubble residuals are calculated fordata with and without CC SNe and differences in distancevs. redshift are used as a systematic uncertainty. We furtherreduce the contamination by a factor of 5 × when consider-ing an additional nearest-neighbor (NN) cut, as described inKessler & Scolnic (2016). We take this reduced contamina-tion as our nominal uncertainty for imaging-only scenarios(see Figure 9c and i). Kessler & Scolnic (2016) reduces thecontamination by a factor of 3.6 × with the NN cut, but weare slightly more optimistic because of the additional rest-frame NIR data and stricter selection cuts. A full simulationwith NN classification is beyond the scope of this work, butwill be considered in future investigations.Our optimistic core-collapse contamination uncertainty for imaging-only simulations is assumed to be negligible, aswe expect classification methods to have improved substan-tially by launch and to be able to take advantage of therest-frame NIR data. Our simulations also take into account five systematicuncertainties related to SN physics. The host-galaxy – SN luminosity relation: After cor-recting for SN light-curve shape and colour, SN Ia Hub-ble residuals still correlate with host-galaxy properties (e.g.,Kelly et al. 2010; Lampeitl et al. 2010; Sullivan et al. 2010).Although the cause of this effect is still unknown, it is possi-ble that it is related to different progenitor properties, suchas metallicity or age, that correlate with environment.Currently, cosmology analyses (e.g., Betoule et al. 2014)correct SN luminosities based on the mass of the SN host-galaxy relative to a central split value. The exact functionalform of this correction is still poorly constrained, but mostassume a binary population split at 10 M (cid:12) (Sullivan et al.2010). It is possible that the magnitude of this correction andthe form could change with redshift (e.g., Rigault et al. 2013,2015; Childress et al. 2014). However, the size of the system-atic uncertainty due to the mass-dependent evolution can bemitigated by measuring the relation between distance resid-uals and mass at different redshifts. This method is similarto ideas presented by Shafer & Huterer (2014). Therefore,the size of the systematic uncertainty is actually dependenton how well the evolution of the relation is measured.To mimic this effect, we take half of our output SNANASN Ia sample, which we call our “high-mass” sample, andintroduce a redshift-dependent offset in the peak brightnessof the SN m B , following m B, shift = m B + 0 . − [0 . × (1 − z )] . (13)We then determine the redshift dependence of the differencein Hubble residuals between our altered “high-mass” sampleand the unaltered “low-mass” sample. The difference in ourrecovered dependence and our input dependence, given inEquation 13, is used as the size of our nominal systematicuncertainty (see Figure 9d and j). For this analysis, weassume that the uncertainty in the difference of the Hubbleresiduals for the two bins is dominated by the distanceuncertainty, rather than uncertainties in the mass estimatesof the host-galaxies. There may be a larger systematic un-certainty related to a population drift of the host-galaxies;however, we choose this particular kind of systematic forthe host-galaxy - SN luminosity relation to represent varioussystematic uncertainties that actually improve with greaterstatistics. As such there is no optimistic value for this biasas it is based purely upon the statistics of the survey. Intrinsic Scatter Uncertainty: There is still uncer-tainty in the relative proportion of colour variation and lu-minosity variation in the intrinsic scatter model for SNe Ia(see Scolnic & Kessler 2016, for a review). The distancebias corrections applied depend on the assumption of theintrinsic scatter model. The differences between the biascorrections are typically largest where selection effects arestrongest because the intrinsic scatter model will determinewhether predominantly bluer objects are selected or pre-dominantly brighter objects are selected. To determine theimpact on our cosmological measurements from this uncer-tainty, we first simulated our samples with two different in-trinsic scatter models: G10, a model from Guy et al. (2010) MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey which has 70% luminosity variation and 30% colour varia-tion, and a model from Chotard et al. (2011), C11, which has25% luminosity variation and 75% colour variation. Follow-ing Kessler et al. (2013), we converted the Guy et al. (2010)and Chotard et al. (2011) models into spectral-variationmodels for SNANA. The difference between the recovereddistances from these two models is the systematic uncer-tainty, shown in Figure 9e. The structure of the distancedifferences with redshift shown in Figure 9e is due to theimpact of various selection effects (from the tiered surveys)on the different scatter models.The optimistic intrinsic scatter uncertainty is assumedto be 5 × better than current estimates due to improvedmodels in the IR (see Mandel et al. 2011) Population drift: Related to uncertainty in the intrin-sic scatter model, there is uncertainty in whether this formof the scatter could evolve with redshift. This issue is con-flated in past analyses with the possibility that the colour ofthe SN population could evolve with redshift (Mandel et al.2016; Scolnic & Kessler 2016), and this evolution is not ac-counted for in the analysis. To determine the impact on ourcosmological measurements from this uncertainty, we intro-duced a SN colour population drift of 0 . × z mag, keepingthe defined colour range and Bifurcated Gaussian σ identicalto previous simulations.While there may be evidence for an x population drift(see Scolnic & Kessler 2016), it will have less impact on pos-sible distance biases than a c population drift because ofthe different correlations between c and x with luminosity.Therefore in this analysis, we do not include an additional x population drift. The difference between the recovered dis-tances from this shift and the nominal simulation are shownin Figure 9f, with relative FoM values given in Figure 9l.It is possible that with the IFC-S, evolution of the in-trinsic colour can be constrained by measuring the SN ejectavelocities (Foley & Kasen 2011; Foley et al. 2011; Foley 2012;Mandel et al. 2014). This claim is analysed further in Ap-pendix B, though for our nominal systematic uncertainty,we do not assume any improvement in the constraint on in-trinsic colour evolution or population drift from the IFC-S.The optimistic population drift uncertainty for ∆ z = 0 . tot predictions, but their effects have beenconsidered. Beta evolution: The properties of interstellar dustmay change with redshift, affecting the ratio of total toselective extinction. This evolution would manifest itself ina change in the recovered value of β (Scolnic et al. 2014a)with redshift (Conley et al. 2011). Furthermore as shown inMandel et al. (2016), β may be composed of a reddeninglaw as well as a separate relation between SN intrinsiccolor with luminosity, and the relative components of thetwo may change with redshift. Similar to the correlationof Hubble residuals with host-mass, β evolution can beincluded as a fit parameter and its uncertainty will decrease with the size of the statistical sample. Therefore, its uncer-tainty is expected to be small compared to the systematicuncertainty from the intrinsic uncertainty or populationdrift, so is not included here. K-corrections: The SDT report lists K-corrections asa top systematic uncertainty and a large motivation for theuse of the IFC-S over broad-band imaging. However, sincemodern distance-fitting algorithms employ spectral modelsto fit SN data in the observer frame, no true K-correction isever applied. Instead, a K-correction uncertainty should bedescribed as an imperfect knowledge of a SN SED. Sincemost of our SN Ia training set is at z ≈ 0, certain re-gions of the spectral model (near the effective wavelengthof certain filters) are better constrained than others. If thede-redshifted observer-frame and rest-frame filters are notwell aligned, the diversity of spectral features could causean additional statistical uncertainty of up to 0.04 mag (e.g.,Saunders et al. 2015). With IFC-S measurements, one cansynthesise photometry over any wavelength range, largelyeliminating this uncertainty.It has been argued (e.g., Aldering et al. 2002; Perl-mutter & Schmidt 2003) that this uncertainty will bedominant for a space-based SN mission. However, evenwith the most pessimistic scenario (Saunders et al. 2015),this uncertainty is still negligible compared to the 0.13 magintrinsic scatter of SN Ia (Scolnic et al. 2014b; Betouleet al. 2014). This bias averages out on redshift scales of∆ z ≈ . 1. To further examine this point, the IFC-S spectrawere binned, maintaining the overall SNR, mimickingprogressively lower resolution spectra (or wider filters).No systematic bias is found and distance uncertaintiesdo not increase, confirming that this uncertainty will besub-dominant. Therefore, K-corrections are not included asan additional systematic uncertainty. Milky Way extinction: Systematic uncertainties inthe amount of Milky Way (MW) extinction in the line-of-sight to the SNe will propagate to systematic uncertainties inthe recovery of the cosmological parameters. The WFIRST SN fields have not yet been chosen, but it is likely theywill be picked to minimize the amount of MW extinction:MW E ( B − V ) < . 02 mag. As discussed in Scolnic et al.(2014a), systematic uncertainties in the MW extinction takethe form of a multiplicative component and additive com-ponent. Assuming a 10% multiplicative uncertainty and aseparate 3 mmag additive uncertainty, we find the impacton the FoM is small ( < A summation of the current and optimistic systematicuncertainties investigated by our various simulations is pre-sented in Table 12.The effect of each individual systematic uncertaintyboth current and optimistic is presented within Figure 11.The values plotted here are produced using CosmoMC*. Forthe SDT survey simulation, the largest uncertainty is the colour-gradient uncertainty. Our current estimate for this MNRAS , 1–29 (2016) R. Hounsell et al. Table 12. Current and optimistic systematic uncertainties investigated for both the WFC and IFC-S. Since the impact of systematicuncertainties such as beta evolution , K-corrections , and MW extinction are considered negligible, we have not included them within ourfinal analysis. Systematic Current OptimisticUncertainty WFC IFC-S WFC IFC-SColour-gradient 5 mmag per 7000 ˚A 50 mmag per 7000 ˚A 3 mmag per 7000 ˚A 3 mmag per 7000 ˚ANon-linearity 1% per dex over 10 mag 1% per dex over 10 mag 0.2% per dex over 10 mag 0.2% per dex over 10 magZero-point offsets 5 mmag · · · · · · CC contamination 1/5th of derived systematic c 0% 0% 0%Population drift 10 mmag × z 10 mmag × z . × z . × z Intrinsic scatter The difference between the G10 and C11 models 1/5th that of current 1/5th that of currentHost-mass evolution Calculated for each strategy d Beta evolution Considered Negligible e K-corrections Considered Negligible e MW extinction Considered Negligible ec See Section 6.2 for details on this systematic uncertainty. d For each simulated survey strategy the host-mass systematic uncertainty was calculated as described within Section 6.3. e As the effect of this systematic uncertainty is considered negligible (see Section 6.3) we have not included it within our final analysis. uncertainty is 50 mmag per 7000 ˚A. We hope that this valuewill decrease by over a factor of 10 × by launch, leading tothe much larger relative FoM tot , opt . For the imaging-onlyscenario the largest systematic uncertainties are the intrin-sic scatter and the zero-point offsets , specifically for the Y and H bands. Further evaluation of these uncertainties is re-quired in order to fully understand their effects and enableoptimisation of survey strategies. Within this paper we have simulated a total of 11 dif-ferent SN survey strategies for the WFIRST mission. Herewe compare each strategy, assessing how successful they areat constraining dark energy models, via their FoM values.We also examine the details of these strategies, such as red-shift distribution of SNe Ia, and suggest how they may beimproved.Using the “optimised” systematic uncertainties de-scribed above we have evaluated the impact of each uncer-tainty on each of our simulated surveys, the results of whichare presented in Figure 12, with Table 13 listing the FoM stat ,FoM tot , curr , and FoM tot , opt values determined for each case.For completeness the FoM values presented here are calcu-lated using the original version of CosmoMC, where the fullset of Planck likelihoods are considered. Figure 12 indicatesthat the surveys examined possess a wide range of FoM stat (103 – 476) values compared to a much narrower range ofFoM tot , curr (71– 198) values.Examination of strategies that use both the IFC-S andWFC (e.g., SDT, SDT* and SDT* Highz) allow us to drawseveral important conclusions. The SDT strategy as outlinedin Spergel et al. (2015) results in a lower than expectednumber of SNe Ia at z < . 6. This decrease in low- z SNeis a result of the short exposure time within the shallowtier of the imaging survey, and the strict spectrophotometricselection criteria.Slight modification of these selection criteria as imple-mented in SDT* (see Figure 7a) increases the total numberof low- z SN Ia by ∼ ∼ 28% over0 . (cid:54) z (cid:54) . 7. The short exposure time of the shallow imag-ing tier leads to low-SNR SNe and thus even with modifiedselection criteria this shallow survey still hinders the num-ber of z < . ∼ 43% of thefraction stated within the SDT report.Based on these results we conclude that any selection criteria implemented must be very carefully chosen so as tomaximise both efficiency and purity, and that the shallowtier of the imaging survey (with current overhead estimates)is a sub-optimal use of survey time. Shifting exposure timefrom the shallow imaging tier to the medium tier and apply-ing the modified selection criteria significantly increases thenumber of SNe Ia observed (475 and 915 more SNe Ia com-pared to SDT*, and SDT surveys, respectively) as indicatedin our SDT* Highz survey simulation (see Figure 7b). Thisstrategy possesses a much higher FoM tot , opt value of 364 incomparison to the FoM tot , opt = 284 value of the SDT.Figure 13 (left) presents the w − w a 68% and 95% con-fidence contours for the simulated SDT, SDT*, and SDT*Highz surveys. These contours illustrate how slight modifica-tion of the classification criteria presented by the SDT report(see Section 3.1) can lead to an increase in the FoM tot , opt whereas moving time to focus on the medium tier of thesurvey makes for a more significant impact.As an informative worst-case scenario, the SDT Imag-ing simulation replicates a situation where all IFC-S data isdetermined to be unusable, but only after completion of the WFIRST mission. As a result, only the imaging data as partof the SDT survey would be used for cosmological analyses.Unsurprisingly, this SN survey produces too few SNe Ia at z < . tot , opt = 78.The Imaging:Allz simulation has a FoM tot , opt = 359,and is one of our more successful imaging-only strategies.It is a 3-tier imaging strategy that uses four broadband fil-ters ( RZY J or Y JHF ). This survey discovers > stat = 456.Combined, Zero-point uncertainties are the largest system-atic uncertainties for this strategy (see Figure 11), as withall imaging strategies.When comparing the Imaging:Allz and SDT* HighzFoM stat values, it is interesting to note that even thoughthe final cosmology sample of the Imaging:Allz survey con-tains 33% more SNe Ia at z > . 2, the addition of these SNeincreases the relative FoM stat value by only ∼ w − w a parameterization.We have generated and examined four Lowz two-tierimaging-only surveys in which time from the IFC-S anddeep tier have been redistributed amongst the shallow andmedium tiers of the discovery survey. This has allowed forthe addition of several filters and an increase in each tier’sobservational area. Each of these simulated surveys stillfailed to meet the required number of SN Ia (as outlined MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey Figure 11. Total (statistical+systematic) FoM values when including only a single systematic uncertainty relative to FoM stat . Eachsystematic uncertainty considered is listed in the figure. The range for each uncertainty represents the possible effect on the FoM if thesystematic uncertainty is not improved from its current state (dotted line) to our optimistic value. The top and bottom panels displaythe systematic uncertainties considered for the SDT (see Section 5.1) and Imaging:Allz strategies (see Section 5.4), respectively. In caseswhere noise fluctuations makes the relative FoM slightly greater than FoM stat , these value have been set to 1. Note that the SDT resultsdo not include the effect of zero-point uncertainties as this strategy does not use any imaging to measure distances. Since the impact ofsystematic uncertainties such as beta evolution , K-corrections , and MW extinction are considered negligible (Section 6.3), we have notincluded them here. Figure 12. Predicted dark energy FoMs for the simulated WFIRST SN survey strategies outlined in Section 5. IFC-focused and WFC-focused strategies are presented in the top and bottom panels, respectively. The gradients for each strategy represent the range of FoMsfrom FoM tot , curr (dotted lines) to FoM tot , opt . The thick black lines represent the FoM stat values of each simulation. The red dashed lineindicates the current FoM value of 32.6 (Alam et al. 2016).MNRAS000 SN survey strategies outlined in Section 5. IFC-focused and WFC-focused strategies are presented in the top and bottom panels, respectively. The gradients for each strategy represent the range of FoMsfrom FoM tot , curr (dotted lines) to FoM tot , opt . The thick black lines represent the FoM stat values of each simulation. The red dashed lineindicates the current FoM value of 32.6 (Alam et al. 2016).MNRAS000 , 1–29 (2016) R. Hounsell et al. Table 13. Statistical and Systematic FoM values and their associated σ for each simulated survey strategy investigated.Strategy f Statistical Current OptimisticFoM σ ( w ) σ ( w a ) FoM σ ( w ) σ ( w a ) FoM σ ( w ) σ ( w a )SDT 318 0.043 0.20 157 0.050 0.23 284 0.044 0.20 SDT* 371 0.037 0.17 174 0.047 0.21 338 0.038 0.17 SDT* Highz 412 0.034 0.15 198 0.045 0.19 364 0.036 0.16SDT Imaging 141 0.081 0.36 71 0.102 0.38 78 0.098 0.38 Imaging:Allz 456 0.030 0.14 195 0.053 0.24 359 0.035 0.17Imaging:Lowz 103 0.083 0.36 82 0.092 0.38 93 0.088 0.37Imaging:Lowz* 375 0.035 0.17 163 0.058 0.28 309 0.038 0.19Imaging:Lowz+ 353 0.038 0.18 158 0.058 0.27 274 0.040 0.19Imaging:Lowz-Blue 334 0.039 0.19 176 0.057 0.26 303 0.040 0.20 Imaging:Highz* 476 0.028 0.14 188 0.052 0.23 369 0.033 0.17Imaging:Highz+ 456 0.029 0.15 166 0.055 0.26 314 0.035 0.17 f The ordering of this table is based upon the ordering of simulations presented in Section 5. Strategy names marked as bold representthe survey simulations with the highest FoM tot , opt values. Figure 13. w − w a 68% and 95% confidence contours for the simulated SDT, SDT*, and SDT* Highz WFIRST SN surveys (leftpanel) and the SDT, Imaging:Allz, and Imaging:Highz* WFIRST SN surveys (right panel). Each contour represents total (statisticalplus optimistic systematic uncertainties) SN Ia constraints combined with CMB and BAO constraints. For comparison we have includedthe confidence contours created using CMB+BAO data only. in the SDT report) at z > . 2. The Imaging:Lowz surveylacks the desired SNe Ia at z < . z > . 1, resulting inthe small FoM tot , opt = 93.For the two Highz imaging-only strategies, time fromIFC-S and the shallow tier observations was re-distributed tothe medium and deep tiers and observations were made withadditional filters. The Imaging:Highz* survey is the mostsuccessful imaging-only survey with FoM tot , opt = 369. It alsohas very small statistical uncertainties with FoM stat = 476,the largest statistical-only FoM for any strategy examined.Figure 12 visually compares the different FoM estimatesfor each strategy. The SDT Imaging and Imaging:Lowzstrategies are clearly less precise than others since theirFoM stat and FoM tot , opt values are below the FoM tot , curr val-ues of all other surveys. These strategies are clearly inferiorto other options.Since the SDT and SDT* strategies are equivalent ex-cept in classification, the final systematic uncertainty foreither strategy would be essentially equivalent. Therefore, itis clear that the SDT* strategy is superior to that of theSDT report. The SDT FoM tot , curr value is also comparableto that possessed by many of the other strategies, an effectthat can be attributed to the fact that at this point we aresystematics limited. For the remaining strategies, it is often difficult to havea clear ranking. The effectiveness of each strategy has dif-ferent dependencies on specific improvements in systematicuncertainties. For instance, if all systematic uncertaintiesimprove significantly except for our ability to calibrate IFUspectrophotometry, all IFC-focused strategies will have aFoM value close to FoM tot , curr while all imaging-only strate-gies would have a FoM value closer to FoM tot , opt .None the less, our current simulations still provide im-portant information about where to focus efforts. Consid-ering the FoM tot , opt values, the top 3 strategies are Imag-ing:Highz*, Imaging:Allz, and SDT* Highz, which all havesimilar FoM tot , opt values. There is no obvious optimal strat-egy among those investigated here and with current knowl-edge. Importantly, imaging-only strategies are capable ofconstraining dark energy as well as IFC-S strategies. Fig-ure 13 (right) presents the w − − w a 68% and 95% con-fidence contours for the simulated SDT, Imaging:Allz, andImaging:Highz* surveys. These contours illustrate how com-petitive imaging-only strategies are with respect to an IFC-focused strategy.The wavelength dependent calibration uncertainty forthe IFC-S system is currently large enough to significantlyhamper the effectiveness of any IFC-focused strategy. We MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey are optimistic that by launch it will improve by a factor of17 (see Figure 11). However since no clear path has been pre-sented for this improvement, we have also investigated howfactors of 5 and 10 improvement (i.e., 25, and 5 mmag per7000 ˚A) affect the final FoM values of the SDT* strategy. Forimprovement factors of 1 (no improvement; current value),5, 10, and 17 (optimistic value), we find FoM tot = 223, 229,308, and 338, respectively. For these calculations, the val-ues of the other systematic uncertainties (i.e., non-linearity , host-mass evolution , population drift , and intrinsic scatter )are set to their optimistic values. It is clear that a preci-sion of at least 5 mmag per 7000 ˚A is required for optimalimplementation of an IFC-focused strategy.In addition imaging-only strategies like Imaging:Allzand Highz* may also have an advantage in that their datacan be sub-divide into samples for further systematic stud-ies. For example, high and low- z host-mass and high andlow- z Galactic extinction studies. If new effects are foundsuch as β ( z ) or a better host-mass function, then imaging-only strategies with superior statistics will prove better formeasuring these additional parameters. The strategies outlined in this paper illustrate how the WFIRST SN survey can be modified to increase the numberof SNe Ia examined, and the redshift range over which theyare found. These strategies are intended as reference optionsthat can be updated and expanded upon to perform morerigorous optimisations.Future optimisation of the survey may include tradingdepth or area and adjusting the cadence of the light curves.In addition, the current redshift distribution proposed bythe SDT report could be further optimised with relativelysmall tweaks to the survey. Below we discuss in more detailsome of the ways in which the survey could be optimised.Our simulations currently assume that the redshift ofeach SN is perfectly known. In reality, the SNe will havevarying levels of redshift accuracy/precision based on howthe redshift is determined. The accuracy of the redshift af-fects observation choices such as exposure times, the accu-racy of classification routines, and potential biases that prop-agate to the Hubble diagram. Meanwhile the uncertainty inthe redshift propagates directly to constraining cosmologicalparameters.We will likely use a combination of relativelyhigh-resolution spectroscopic host-galaxy redshifts, lower-resolution WFIRST grism host-galaxy redshifts, SN+galaxyphotometric redshifts, and spectroscopic redshifts from theSNe themselves. Further complicating the issue, the redshifts(and their uncertainties) will be updated and improved dur-ing the course of the survey.A full analysis of these effects requires an accurate as-sessment of the redshift catalogs present at the beginningof the SN survey, the ground-based resources available dur-ing the survey, the exact WFIRST survey strategy, and re-sources available upon completion of the survey. With esti-mates of the available resources, we can assign redshifts withappropriate accuracy to each simulated SN and determinehow each survey is affected.Our simulations have followed the current NASA man- date that all SN discovery and follow-up observations beperformed exclusively by WFIRST . However, we will likelyobserve WFIRST -discovered SNe from the ground. Further-more, it is possible for WFIRST to observe SNe discoveredwith other telescopes. There could be significant efficiencygains if one could, for example, follow Large Synoptic Sur-vey Telescope (LSST, Ivezic et al. 2008) discovered SNe Iawith the IFC-S or obtain SN classifications using 8 to 30-mclass telescopes. However, ground-based discoveries have ad-ditional potential systematic uncertainties because of vari-able seeing, extended periods of bad weather affecting ca-dence, etc. Future simulations should examine the possibilityof multiple scenarios for using ground-based observatories toenhance the WFIRST SN survey.The use of the grism has not yet been fully explored orsimulated. We performed preliminary simulations, findingthat grism spectroscopy would be effective for classification,but only with longer exposure times (see Jones et al. 2013,for a detailed examination of the HST WFC3 grism for thispurpose). More detailed simulations are necessary to deter-mine if the grism is useful for the SN survey.Usage of both the IFC-S and WFC imaging compo-nents with their current 5 day cadence will place consid-erable strain on scheduling. Within 5 days data will have tobe down-loaded, processed, searched for transients, objectsfit and selected, IFC-S follow-up observation schedules builtand sent, and finally the instrument set to observing again.This cadence places a significant strain on human resourcesand is a risk to the mission. A longer cadence of 7 days,or even a flexible cadence may well ease some of these is-sues and have little to no scientific impact on the mission.Modified cadence investigations will be the subject of futurework.The idea of using parallel observing for the IFC-S andWFC must also be considered. Parallel observing would al-low WFIRST to operate both the WFC and IFC-S at thesame time. Preliminary calculations suggest that given the0.477 deg offset from the IFC-S to the center of the WFC,and a random angle, the fraction of random parallel fieldsthat fall into what would be considered the deep SN fieldis ∼ ∼ z > . ∼ 75 IFC-S observations in the field per visit.This means that on average ∼ MNRAS000 75 IFC-S observations in the field per visit.This means that on average ∼ MNRAS000 , 1–29 (2016) R. Hounsell et al. mapping of detector to the sky, out-of-band stray light, etc.Initial assessments show that these calibration uncertaintiesare all second-order systematics that are significantly be-low the ones included in our current analysis, but will bereviewed in future analyses.An important limitation of the accuracy of the SN dis-tances is the training sample used to determine the underly-ing spectral model. As described in Astier et al. (2014), theSN model uncertainty can be reduced by using the samerest-frame wavelength range at all redshifts. For a rest-frame wavelength range of 2800 – 8000 ˚A correspondingto the current SALT2 spectral model, the mean effectivewavelength for ZY JHF filters will fall in redshift ranges of0 . < z < . 1, 0 . < z < . 9, 0 . < z < . 6, 1 . < z < . . < z < . 7, respectively. Extending the SN cosmol-ogy sample to lower redshifts requires further extension ofthe spectral model into the NIR. One likely possibility isthat additional filter slot will be available for a bluer filterso that the NIR extension may not be as critical, but thiswork will still prove beneficial for the program.There have been several efforts to obtain NIR SN Iadata (e.g., Krisciunas et al. 2004; Wood-Vasey et al. 2008;Stritzinger et al. 2011; Friedman et al. 2015). In total, thereare now several hundred SNe Ia with NIR light curves. Whilemost of these data are for low- z SNe, the RAISIN project(PI Kirshner) has collected rest-frame NIR data of ∼ . < z < . 6) SNe Ia with HST /WFC3. HST . In addition to contributing to the NIR model, thesedata will be very useful for investigating systematic uncer-tainties related to intrinsic scatter, dust, and color.When comparing multiple survey strategies, it is bestto have a single, pre-defined metric by which one can com-pare. With multiple metrics, one can generally choose themetric that is optimal for a particular strategy. That said,there can be critical aspects of a problem that do not affecta metric. For instance, the DETF FoM that we use to com-pare strategies does not contain any information related tomission cost/risk or enabling ancillary science.Furthermore, the DETF FoM is not the only metric bywhich we can optimise our understanding of dark energy.For instance, eigenvectors have been a popular approach,Huterer & Starkman (2003) (although Linder & Huterer2005, argue that something like the DETF FoM is sufficientfor most needs). It will be straight-forward to implementsuch dark energy characterisations in our simulations, butinterpretation (and evaluation) will likely be debated.The SN survey defined within the SDT report limits thenumber of SN Ia at high- z and focuses on achieving a largersample within 0 . (cid:54) z (cid:54) . 6. Our imaging-only strategieshowever, place no limit on the number of SN Ia within agiven redshift bin, and explore out to z (cid:54) . 0. As our surveyshave not been optimised, we have not specifically consideredthe effects of focusing observations within any given redshiftrange. However, our preliminary studies have indicated (seeSection 7) that an increase in the fraction of SN Ia withhigher redshifts (i.e., z > . 2) does not necessarily providea significant increase in a surveys FoM. This is likely due tothe nature of dark energy and the w − w a parameterization.Variations on the redshift distribution will be considered aspart of our future optimisation studies.Our work has used constraints on the cosmological pa-rameters from both the BAO (Anderson et al. 2014) and CMB (Planck Collaboration et al. 2016) datasets. However,there is ongoing work to include external constraints fromprojections of CMB S4 (Abazajian et al. 2016) and futureBAO missions, e.g., the DESI Collaboration et al. (2016).Weinberg et al. (2013) showed that the impact of Stage 4SN constraints on the FoM has a strong dependence on therelative constraints from the Stage 4 BAO and future WeakLensing probes. Future work will attempt to replicate thisanalysis in the context of various SN strategies. Using open-source tools, including newly createdones , we have produced the first fully simulated reali-sations of the WFIRST SN survey. We have examined 11strategies in detail, including the survey strategy presentedin the SDT report. For each simulated SN survey strategy,several statistical and systematic uncertainties have beenexamined and included in order to calculate the FoM tot , opt value, which we have used as our final measure of success.Examination of the results produced by our SDT sim-ulation (see Section 5.1) shows that this strategy results infewer SNe Ia than outlined in the SDT report. The selectionefficiency of the SDT strategy is low, and the noise is signif-icantly underestimated, resulting in many SNe Ia being cutor misclassified in the final sample by the strict classifica-tion routine outlined in Section 3.1. With FoM tot , opt = 284this is, on comparison, one of the least successful surveystrategies investigated (see Figure 12 and Table 13). Modi-fication of the selection criteria to allow for variation in therise between epochs (required due to noise fluctuations), andrestriction of colours (to prevent the inclusion of more ex-otic events), increased this FoM tot , opt to 338. Even withinthe SDT* simulation however, there are still too few SN Iaselected at z < . 6, due in part to the short 13 secondexposure of the shallow imaging tier. Time devoted to theshallow tier of the SDT* survey is placed into the mediumtier to produce our SDT* Highz scenario. For this strategy,FoM tot , opt = 364.Our imaging-only scenarios consist of strategies withonly the shallow+medium tiers (suffix Lowz), all threeimaging tiers (SDT Imaging and Imaging:Allz), and themedium+deep tiers (suffix Highz). Tier areas were increasedand additional filters added to compensate for the loss of theIFC-S and/or discovery tiers. Additional filters allowed abroader coverage of the rest-frame optical wavelengths (viaaddition of R and Z -bands), a region where our spectralmodels are well defined, and extension into the rest-frameNIR (via addition of F -band). The Imaging:Allz and Imag-ing:Highz* surveys have FoM tot , opt values of 359 and 369respectively, making them comparable to the SDT* Highzstrategy. The Imaging:Highz* survey also has the highestFoM stat value at 476. For many of the imaging-only strate-gies the number of SNe Ia within the final sample is signifi-cantly higher than that obtained by the SDT strategy.Using the FoM tot to measure the success of each strat-egy, there is no best strategy. The Imaging:Highz*, SDT*Highz, and Imaging:Allz simulated surveys all have similar See https://jet.uchicago.edu/blogs/WFIRST/ MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey current and optimistic FoM values. Until systematic uncer-tainties are further constrained, we cannot say if using theIFC-S is a net benefit to the WFIRST SN survey.However, there are several additional concerns relatedto the use of the IFC-S that must be addressed before anIFC-focused strategy can be adopted. Specifically, an IFC-focused strategy must have active target selection (likelywith human decisions included), which increases operationscosts and locks in selection bias at the time of target selec-tion. The ability to produce high-precision spectrophotom-etry with an IFC has yet to be demonstrated, resulting ina higher risk of reaching systematic uncertainty goals thanfor an imaging-only strategy. An IFC-focused approach thatrequires both the imager and IFC-S also increases risk of afatal hardware failure over a strategy that uses only one in-strument. A further limitation of IFC-focused strategies aretheir relatively small sample sizes that will prevent somestudies that require subdividing the sample into relativelysmall parameter-space bins.While the strategies we have presented are not fullyoptimised, they provide a broader understanding of the pos-sibilities for the survey. Moreover, at this stage in the mis-sion, such an investigation is critical for mitigating risk andensuring the ultimate success of WFIRST . Our initial in-vestigations have determined that there is no one correctsurvey scenario for the mission, yet all our top strategieswill provide a significant improvement in comparison to cur-rent surveys which utilize SNe Ia as cosmological probes,and progression towards that which is expected by a Stage4 experiment. Our work, which has focused on developingand establishing a reliable and reproducible set of baselinestrategies, will enable future optimization of the survey viaapplication of the factors and tools mentioned within ourdiscussion, and thus produce a more definitive and success-ful survey strategy. ACKNOWLEDGEMENTS This manuscript is based upon work supported by theNational Aeronautics and Space Administration under Con-tract No. NNG16PJ34C issued through the WFIRST Sci-ence Investigation Teams Program.It was also supported in part by the U.S. Departmentof Energy under contract DE–AC02–76CH03000. Analysiswas done using the Midway-RCC computing cluster at Uni-versity of Chicago.R.H., D.S., and R.J.F. were supported in part by NASAgrant 14–WPS14–0048. The UCSC group is supported inpart by fellowships from the Alfred P. Sloan Foundation andthe David and Lucile Packard Foundation to R.J.F.R.H. would like to thank P. Szuta for his help and sup-port during the completion of this paper.D.S. and R.K. acknowledge support from the Kavli In-stitute for Cosmological Physics at the University of Chicagothrough grant NSF PHY–1125897 and an endowment fromthe Kavli Foundation and its founder Fred Kavli. D.S. isalso supported by NASA through Hubble Fellowship grantHST–HF2–51383.001 awarded by the Space Telescope Sci-ence Institute, which is operated by the Association of Uni-versities for Research in Astronomy, Inc., for NASA, undercontract NAS 5-26555, and is a Hubble/KICP Fellow. V.M. was supported in part by the Charles E. Kauf-man Foundation, a supporting organization of the Pitts-burgh Foundation.Supernova cosmology at the Harvard College Observa-tory is supported in part by the National Science Foun-dation through grants AST–156854, AST– 1211196, andNASA grant NNX15AJ55G. R.P.K. was supported in partby the RAISIN1 GO–13046 grant and the RAISIN2 GO–14216 grant, which were administered by the Space Tele-scope Science Institute, which is operated by the Associa-tion of Universities for Research in Astronomy, Inc., underNASA contract NAS 5–26555.We would also like to thank A. Friedman for providinghis compilation of public NIR photometry. REFERENCES Abazajian K. N., et al., 2016, preprint, ( arXiv:1610.02743 )Alam S., et al., 2016, preprint, ( arXiv:1607.03155 )Albrecht A., et al., 2006, preprint, ( arXiv:astro-ph/0609591 )Aldering G., 2001, LBNL, 51157, 1Aldering G., et al., 2002, in J. A. Tyson & S. Wolff ed., Soci-ety of Photo-Optical Instrumentation Engineers (SPIE) Con-ference Series Vol. 4836, Society of Photo-Optical Instru-mentation Engineers (SPIE) Conference Series. pp 61–72,doi:10.1117/12.458107Anderson L., et al., 2014, MNRAS, 441, 24Astier P., et al., 2014, A&A, 572, A80Bacon R., et al., 2015, A&A, 575, A75Bailey S., et al., 2009, A&A, 500, L17Bernstein J. P., et al., 2012, ApJ, 753, 152Betoule M., et al., 2013, A&A, 552, A124Betoule M., et al., 2014, A&A, 568, A22Blondin S., Tonry J. L., 2007, ApJ, 666, 1024Blondin S., Mandel K. S., Kirshner R. P., 2011, A&A, 526, A81+Bohlin R. C., 2007, in Sterken C., ed., Astronomical Society of thePacific Conference Series Vol. 364, The Future of Photometric,Spectrophotometric and Polarimetric Standardization. p. 315( arXiv:astro-ph/0608715 )Bohlin R. C., Gordon K. D., Tremblay P.-E., 2014, PASP, 126,711Cardelli J. A., Clayton G. C., Mathis J. S., 1989, ApJ, 345, 245Chevallier M., Polarski D., 2001, International Journal of ModernPhysics D, 10, 213Childress M. J., Wolf C., Zahid H. J., 2014, MNRAS, 445, 1898Childress M. J., et al., 2016, Publ. Astron. Soc. Australia, 33,e055Chotard N., et al., 2011, A&A, 529, L4+Conley A., et al., 2011, ApJS, 192, 1Content D., et al., 2013, in UV/Optical/IR Space Telescopes andInstruments: Innovative Technologies and Concepts VI. pp88600E–88600E–9, doi:10.1117/12.2027717, http://dx.doi.org/10.1117/12.2027717 Contreras C., et al., 2010, AJ, 139, 519DES Collaboration 2005, astro-ph/0510346,DESI Collaboration et al., 2016, preprint, ( arXiv:1611.00036 )Efstathiou G., Sutherland W. J., Maddox S. J., 1990, Nature,348, 705Fakhouri H. K., et al., 2015, ApJ, 815, 58Folatelli G., et al., 2010, AJ, 139, 120Foley R. J., 2012, ApJ, 748, 127Foley R. J., 2013, MNRAS, 435, 273Foley R. J., Kasen D., 2011, ApJ, 729, 55Foley R. J., Sanders N. E., Kirshner R. P., 2011, ApJ, 742, 89Foley R. J., et al., 2012, AJ, 143, 113MNRAS000 Contreras C., et al., 2010, AJ, 139, 519DES Collaboration 2005, astro-ph/0510346,DESI Collaboration et al., 2016, preprint, ( arXiv:1611.00036 )Efstathiou G., Sutherland W. J., Maddox S. J., 1990, Nature,348, 705Fakhouri H. K., et al., 2015, ApJ, 815, 58Folatelli G., et al., 2010, AJ, 139, 120Foley R. J., 2012, ApJ, 748, 127Foley R. J., 2013, MNRAS, 435, 273Foley R. J., Kasen D., 2011, ApJ, 729, 55Foley R. J., Sanders N. E., Kirshner R. P., 2011, ApJ, 742, 89Foley R. J., et al., 2012, AJ, 143, 113MNRAS000 , 1–29 (2016) R. Hounsell et al. Friedman A. S., et al., 2015, ApJS, 220, 9Frieman J. A., Huterer D., Linder E. V., Turner M. S., 2003,Phys. Rev. D, 67, 083505Frieman J. A., Turner M. S., Huterer D., 2008a, ARA&A, 46, 385Frieman J. A., et al., 2008b, AJ, 135, 338Ganeshalingam M., et al., 2010, ApJS, 190, 418Ganeshalingam M., Li W., Filippenko A. V., 2013, MNRAS,Graur O., et al., 2014, ApJ, 783, 28Grogin N. A., et al., 2011, ApJS, 197, 35Gunnarsson C., Dahl´en T., Goobar A., J¨onsson J., M¨ortsell E.,2006, ApJ, 640, 417Guy J., et al., 2007, A&A, 466, 11Guy J., et al., 2010, A&A, 523, A7+Hamuy M., et al., 2006, PASP, 118, 2Hicken M., et al., 2009a, ApJ, 700, 331Hicken M., Wood-Vasey W. M., Blondin S., Challis P., Jha S.,Kelly P. L., Rest A., Kirshner R. P., 2009b, ApJ, 700, 1097Hicken M., et al., 2012, ApJS, 200, 12Hirata C. and Penny M., 2014, White Paper, 3, 1Holz D. E., Hughes S. A., 2005, ApJ, 629, 15Hsiao E. Y., Conley A., Howell D. A., Sullivan M., Pritchet C. J.,Carlberg R. G., Nugent P. E., Phillips M. M., 2007, ApJ, 663,1187Huterer D., Starkman G., 2003, Physical Review Letters, 90,031301Ivezic Z., et al., 2008, preprint, ( arXiv:0805.2366 )Jha S., et al., 2006, AJ, 131, 527Jones D. O., et al., 2013, ApJ, 768, 166Jones D. O., et al., 2016, preprint, ( arXiv:1611.07042 )J¨onsson J., et al., 2010, MNRAS, 405, 535Joyce A., Lombriser L., Schmidt F., 2016, Annual Review of Nu-clear and Particle Science, 66, 95Kelly P. L., Hicken M., Burke D. L., Mandel K. S., Kirshner R. P.,2010, ApJ, 715, 743Kessler R., Scolnic D., 2016, preprint, ( arXiv:1610.04677 )Kessler R., et al., 2009a, PASP, 121, 1028Kessler R., et al., 2009b, ApJS, 185, 32Kessler R., et al., 2010, PASP, 122, 1415Kessler R., et al., 2013, ApJ, 764, 48Kim A. G., Linder E. V., Miquel R., Mostek N., 2004, MNRAS,347, 909Koekemoer A. M., et al., 2011, ApJS, 197, 36Krisciunas K., et al., 2004, AJ, 128, 3034Lampeitl H., et al., 2010, ApJ, 722, 566Lewis A., 2013, Phys. Rev. D, 87, 103529Li W., et al., 2011, MNRAS, 412, 1441Linder E. V., 2003, Physical Review Letters, 90, 091301Linder E. V., Huterer D., 2005, Phys. Rev. D, 72, 043509Maguire K., et al., 2012, MNRAS, 426, 2359Mandel K. S., Narayan G., Kirshner R. P., 2011, ApJ, 731, 120Mandel K. S., Foley R. J., Kirshner R. P., 2014, ApJ, 797, 75Mandel K. S., Scolnic D., Shariff H., Foley R. J., Kirshner R. P.,2016, preprint, ( arXiv:1609.04470 )Miknaitis G., et al., 2007, ApJ, 666, 674Milne P. A., Foley R. J., Brown P. J., Narayan G., 2015, ApJ,803, 20Narayan G., et al., 2016, ApJS, 224, 3National Research Council 2010, New Worlds, New Horizons inAstronomy and Astrophysics. The National Academies Press,Washington, DC, doi:10.17226/12951Peebles P. J. E., 1984, ApJ, 284, 439Perlmutter S., Schmidt B. P., 2003, in Weiler K., ed., LectureNotes in Physics, Berlin Springer Verlag Vol. 598, Supernovaeand Gamma-Ray Bursters. pp 195–217Perlmutter S., et al., 1999, ApJ, 517, 565Planck Collaboration et al., 2016, A&A, 594, A14Rauscher B. J., et al., 2007, PASP, 119, 768Rest A., et al., 2014, ApJ, 795, 44 Riess A., 2010, in Hubble after SM4. Preparing JWST. p. 28Riess A. G., et al., 1998, AJ, 116, 1009Riess A. G., et al., 1999, AJ, 117, 707Riess A. G., et al., 2007, ApJ, 659, 98Rigault M., et al., 2013, A&A, 560, A66Rigault M., et al., 2015, ApJ, 802, 20Rodney S. A., et al., 2012, ApJ, 746, 5Rodney S. A., et al., 2014, AJ, 148, 13Rubin D., Hayden B., 2016, ApJ, 833, L30Rubin D., et al., 2015, ApJ, 813, 137Sako M., et al., 2014a, preprint, ( arXiv:1401.3317 )Sako M., et al., 2014b, preprint, ( arXiv:1401.3317 )Saunders C., et al., 2015, ApJ, 800, 57Schlafly E. F., Finkbeiner D. P., 2011, ApJ, 737, 103Scolnic D., Kessler R., 2016, ApJ, 822, L35Scolnic D. M., Riess A. G., Foley R. J., Rest A., Rodney S. A.,Brout D. J., Jones D. O., 2014a, ApJ, 780, 37Scolnic D., et al., 2014b, ApJ, 795, 45Scolnic D., et al., 2015, ApJ, 815, 117Shafer D. L., Huterer D., 2014, Phys. Rev. D, 89, 063510Spergel D., et al., 2013, preprint, ( arXiv:1305.5425 )Spergel D., et al., 2015, preprint, ( arXiv:1503.03757 )Stritzinger M. D., et al., 2011, AJ, 142, 156Strolger L.-G., et al., 2015, ApJ, 813, 93Sullivan M., et al., 2010, MNRAS, 406, 782Sullivan M., et al., 2011, ApJ, 737, 102Suzuki N., et al., 2012, ApJ, 746, 85Tripp R., 1998, A&A, 331, 815Weinberg S., 1989, Reviews of Modern Physics, 61, 1Weinberg D. H., Mortonson M. J., Eisenstein D. J., Hirata C.,Riess A. G., Rozo E., 2013, Phys. Rep., 530, 87Wood-Vasey W. M., et al., 2007, ApJ, 666, 694Wood-Vasey W. M., et al., 2008, ApJ, 689, 377 APPENDIX A: IFC-S EXTENDED TABLE: Table A1 presented within this appendix provides acomplete listing of each IFC-S bin used within our simu-lations. A machine readable version of this table is availableonline. This table has been truncated for the archive, a fullversion will be available at the submitted journal or by un-commenting the lines within this source code. APPENDIX B: MEASURING THEPOPULATION DRIFT: The evolutionary change of intrinsic colour for a fixedlight curve shape has been shown to occur at some level(currently poorly constrained) by several studies (Foley et al.2012; Maguire et al. 2012; Milne et al. 2015). We have shownthat the colour difference is empirically correlated with SNejecta velocity (Foley & Kasen 2011; Foley et al. 2011; Foley2012; Mandel et al. 2014). Measuring the ejecta velocity bothremoves this potential bias and improves the distance preci-sion for any given SN. Since the colour change is restrictedto λ < WFIRST redshift range will be affected by different amounts.Although we could exclude all data blue-ward of 4500 ˚Athis would result in using only 35% of the pixels for a z = 1 . > 75 if the data are of high quality MNRAS , 1–29 (2016) imulations of the WFIRST SN Survey Table A1. Each IFC-S bin between 0.42 -2.1 µm . Minimum and maximum wavelength ranges for each bin are give along with theFWHM in pixels and associated sources of noise.Maximum Minimum PSF Zodiacal Noise Thermal NoiseWavelength Wavelength FWHM (e − s − pixel − ) (e − s − pixel − )(˚A) (˚A) (pixels)4200.00 4209.35 1.540 0.000 0.0004209.35 4218.76 1.541 0.000 0.0004218.76 4228.23 1.542 0.000 0.0004228.23 4237.76 1.543 0.000 0.0004237.76 4247.35 1.543 0.000 0.0004247.35 4257.00 1.544 0.000 0.0004257.00 4266.71 1.545 0.000 0.0004266.71 4276.49 1.546 0.000 0.0004276.49 4286.34 1.546 0.000 0.0004286.34 4296.24 1.547 0.000 0.0004296.24 4306.22 1.548 0.000 0.0004306.22 4316.26 1.549 0.000 0.0004316.26 4326.37 1.550 0.000 0.0004326.37 4336.56 1.551 0.000 0.0004336.56 4346.81 1.551 0.000 0.0004346.81 4357.13 1.552 0.000 0.0004357.13 4367.53 1.553 0.000 0.0004367.53 4378.00 1.554 0.000 0.0004378.00 4388.55 1.555 0.000 0.000 (Foley 2013). In fact, we measured a Si II velocity for a z = 1 . 55 SN Ia (Rodney et al. 2012) with an R = 130spectrum (Foley 2013).The most important feature for measuring the ejectavelocity is Si ii λ ∼ z > . 2. To determine if we can measure the ejecta velocitywith realistic IFC-S data, we measured the Si II velocity forall long-exposure spectra in the final SDT sample. Doingthis, we found that the typical velocity uncertainty will be1000 km s − , with a ∼ 5% failure rate. The ejecta velocity arealso biased low by ∼ 500 km s − , although presumably thatcan be corrected with measurements from higher-resolutionspectra (perhaps from the ground).This large velocity uncertainty propagates into a 0.10mag distance modulus uncertainty (Foley et al. 2011). Thisrelatively large uncertainty (as large as the total distanceuncertainty) is caused by a combination of low resolutionand low SNR of the IFC-S spectra. For instance, at infiniteSNR, we find a scatter of 340 km s − (close to the limitfrom galactic rotation) and a bias of 180 km s − . For thegrism, the uncertainty decreases to 800 km s − for the same(binned) SNR as the long IFC-S spectrum, indicating thatmost of the uncertainty is caused by the low SNR.This shows that WFIRST has the potential to measurea SN ejecta velocity, but for this velocity to be helpful forimproving distance estimates, we have found that we re-quire (a true) SNR > 20, beyond the current SDT design.However, a slight modification to the survey design and/orstrategy (higher resolution and/or a higher SNR spectrum)would alleviate this problem while simultaneously improv-ing spectral classification. Although this would require addi-tional exposure time per SN, it would reduce the statisticaluncertainty of each SN.Other studies have indicated that other spectral fea-tures, including flux ratios, can improve distance measure- ments slightly (Bailey et al. 2009; Blondin et al. 2011). Us-ing the simulated long spectra, we measure these flux ra-tios, finding that the uncertainties are generally 20%, whichpropagates into a ∼ This paper has been typeset from a TEX/L A TEX file prepared bythe author.MNRAS000