N- body- spectro- photometric simula4ons for DES and DESpec Carlos Cunha DESpec Mee2ng, Chicago May 3, 212
N- body + galaxy + photometry The DES Simulated Sky + Blind Cosmology Challenge Status Update simulated sky surveys developed with Michael Busha (galaxies + sim) Matt Becker (lensing + sim) Brandon Erickson (sim pipeline) Gus Evrard Andrey Kravtsov DES Mock Pipeline / R. Wechsler Can generate new catalogs in ~1 week. Peter Behroozi (halos) Joerg Dietrich (shapes) Basilio Santiago (stars) Molly Swanson (mask) Eli Rykoff, Rachel Reddick (testing) + additional feedback by CWG, Sarah Hansen, Jiangang Hao, Alex Ji, Eusebio Sanchez, Tim Eifler, Joanne Cohn, Martin White + many, many folks who will do analysis! atoms 4% dark matter 22% dark energy 74% Risa Wechsler Stanford/SLAC/KIPAC Carlos Cunha, Stanford University
K- correct Blanton & Roweis (astro- ph/6617) Cornerstone of the photometric and spectroscopic components of the simula2ons. Used to generate colors and spectra of simulated galaxies. Key issues: galaxy templates and priors (put in separately).
Kcorrect templates 5 eigentemplates obtained using Non- nega2ve Matrix Factoriza2on (NMF). Generated from combina2on of 45 star emission history templates from Bruzual & Charlot (23) + 35 templates from Kewley et al (21). Resolu2on 3 Å from 32 to 95 Å Template- resolu2on: 3 km/s; R=1. Training sets: Spectroscopic: 1,6 SDSS Main sample + 4 LRGs. Photometric: 18, from SDSS Main and LRGs; GOODS; DEEP2; and GALEX
Photometric indicators N(z): redshi` distribu2on for BOSS r<21.8 sample using weights Simula4on: DES simula2ons Error bars: simulated sample variance + shot noise of training sets Sheldon, Cunha, Mandelbaum, Brinkmann, Weaver (211)
Spectroscopic indicators Black lines: SDSS spectra Red lines: Best- fit Kcorrect templates Fig. 8. Best fit model spectra based on the five template fit to g, r and i fluxes, compared to the original SDSS spectra from which we computed those fluxes. The models and the original spectra agree very well. Blanton & Roweis (27)
A spectroscopic simula4on Cunha, Huterer, Lin, Busha, Wechsler et al, any minute now. Paberned a`er VVDS: - 8m telescope - 16,2 secs integra2on - With somewhat higher resolu2on: 7.14 Å/pixel - Spectrograph window: ~56-93 Å - i<24 sample selected from DES simula2ons. Redshi`s derived using rvcsao.xcsao Fourier cross- correla2on algorithm. Transmission fraction Flux [photons/s/nm/arcsec 2 /m 2 ] 35 3 25 2 15 1 5 1.8.6.4.2 Sky background emission 4 5 6 7 8 9 1 11 Wavelength (Angstroms) Atmospheric Instrumental 4 5 6 7 8 9 1 11 Wavelength (Angstroms)
A simula4on: completeness 1 1 Cunha, Huterer, Lin, Busha, Wechsler et al, any minute now. 1.8.8.8 True SSR.6.4.2 162s 486s fiducial original comb2 fiducial.6.4.2 162s 486s.6.4.2 162s 486s.2.4.6.8 1 1.2 1.4 1.6 1.8 2 Redshift 2 2.5 21 21.5 22 22.5 23 23.5 24 i-mag 2 4 6 8 1 12 14 R (redshift confidence) SSR: Spectroscopic Success Rate True SSR: frac2on of galaxies with correct redshi`s. R: Strength of correla2on between observed spectra and best- fifng spectrum in a template library.
.4 1-4.4 A simula4on: completeness.4.8 1.2 1.6 2 z true.4.8 1.2 z true Figure 4. Leakage matrices (P (z spec z true ))forthetrainingsetsselectedbythecu panel). The spectroscopic redshifts were calculated using 16,2 secs exposures with t SSR: Spectroscopic Success Rate 3 2.5 2.5 2 True SSR True SSR 1.8.8 True SSR: frac2on of galaxies with correct redshi`s. r-i r-i 1.5 1.5 1.5.5.6.6.4.4 N(z) Observed SSR: Frac2on of galaxies with redshi` confidence above some threshold (R>6). R: Strength of correla2on between observed spectra and best- fifng spectrum in a template library. -.5 -.5-1 -1 2 2.5 21 21.5 22 22.5 23 23.5 24 2 2.5 21 21.5 22 22.5 23 23.5 24 i-mag i-mag Observed SSR (R > 6.) 3 2.5 2.2.2 1.8 N(z) 1.5.6 r-i 1.5.4.2 -.5-1 2 2.5 21 21.5 22 22.5 23 23.5 24 i-mag Figure 2. Top panel: True spectroscopic success rate (SSR T ), defined as fraction of correct redshifts as a function of true redshift. Rightpanel: Observed SSR (SSR O ), defined as fraction of galaxies with correlation R 6.. Both results assume 162 secs of integration time with the 3 additional templates. Fig area and reds sam 5.2
Spectroscopic failures (wrong redshi`s) Issues: When spec- z s are wrong, they re really wrong. A small speck of wrong redshi`s is enough to mess up cosmological constraints. z spec 2 1.6 1.2.8.4 R > 5. 1 1-1 1-2 1-3 1-4 Sample used in the plot has 98.6% correct redshi`s and cons2tutes 6% of total sample..4.8 1.2 1.6 2 z true Case study: Simula2ons of DES photometry + VVDS- like spec- z s R: cross- correla2on parameter (measures redshi` confidence) Cunha, Huterer, Lin, Busha, Wechsler et al, in prep.
Conclusions N- body + photometric simula2ons improving constantly. Probably preby good for DESpec depths. Spectroscopic simula2ons. First step has been taken. Is current resolu2on of templates (3 km/s) sufficient? Need larger training samples to avoid surprises (Yip et al suggest 1 5 galaxies) And perhaps larger eigenbasis. NMF seems like a convenient tool for building a representa2ve eigenbasis. Use more realis2c noise model with varying observing condi2ons, CCD fringing, etc.
Spectroscopic selec4on issues Issues: Spectroscopic samples are very incomplete Redshi` desert is main issue. Need to apply spectroscopic selec2on to photometric sample. Can do this using neural networks (also seen in Soumagnac et al, in prep.) Case study: Simula2ons of DES photometry + VVDS- like spec- z s True SSR 1.8.6.4.2 4.5 h exposures 8- m telescope.2.4.6.8 1 1.2 1.4 1.6 1.8 2 Redshift True SSR: frac2on of galaxies with correct redshi`s. Cunha, Huterer, Lin, Busha, Wechsler et al, in prep.
Q est : redshi` confidence es2mated with neural net. SSR T : Percentage of correct redshi`s in training sample. z true : bias due to selec2on matching with neural networks: is negligible z spec : bias due to selec2on matching + wrong redshi`s: is substan4al 162 secs bias(w) Selection Gal. Frac. SSR T (%) σ(w) z true z spec Q est > 1.5.75 91.4.7.4 -.52 Q est > 2.5.59 97.8.9.2 -.13 Q est > 3.5.46 99.6.1 -.1 -.2 486 secs Shear- Shear constraints on w Q est > 1.5.96 93.6.6.4 -.39 Q est > 2.5.81 97.8.7.5 -.15 Q est > 3.5.66 99.6.8.3 -.3 Table 2. Statistical and systematical errors in w for the different samples. The bias results shown used the template-fitting photo-zs. The Galax.Frac. column indicates the fraction of galaxies from the full data set that passed the selection cut. Cunha, Huterer, Lin, Busha, Wechsler et al, in prep.
Conclusions Incompleteness: Does not introduce cosmological biases if selec2on matching is performed. Sta2s2cal constraints suffer with reduc2on of sample size. Wrong redshi`s: Cause severe biases. Need beber than 99% correct redshi`s. If 99% accuracy not possible, need to calibrate spectroscopic error distribu2on P(z true z spec ) with deeper sample/beber instrument. Moral of the story: Focus has to be on accuracy of derived redshi`s.
Need spectra, so what? Good spectroscopic samples are hard to come by. Issues Selec4on in observables: typically have many more bright samples than faint samples. Selec4on in non- observables: sample selected for a different purpose with different bands (e.g. DEEP2 survey). Shot- noise: samples are small. Sample variance: surveys are pencil- beam. Spectroscopic failures: Can t get spectra for certain galaxies. Wrong spectroscopic redshi`s.
Outline N- body simula2ons + galaxies + photometry Galaxy spectra The role of Kcorrect An example: simula2ng VVDS What do we need for DESpec
Need spectra, so what? Good spectroscopic samples are hard to come by. Solu2ons Selec4on in observables: e.g. Weights (Lima et al 28) Selec4on in non- observables: Don t do it. Shot- noise: need many galaxies Sample variance: need lots of area. Spectroscopic failures: Can t get spectra for certain galaxies. Wrong spectroscopic redshi`s. Cunha, Huterer, Busha, Wechsler 212 Cunha, Huterer, Lin, Busha, Wechsler et al, in prep.
Selec4on matching with neural net Have a redshi` confidence (Q) for galaxies in spectroscopic sample. Use neural net to find a rela2on between Q and observables (magnitudes). This is Q est. Q est can be calculated for all galaxies in the spectroscopic and photometric samples. Poten2al confusion: Q is a new quality parameter I invented to more closely approximate the quality es2mates of real surveys like VVDS and DEEP2. It s just a rescaling (plus discre2za2on) of the R (cross- correla2on strength) parameter.
Spectroscopic failures (wrong redshi`s) Issues: When spec- z s are wrong, they re really wrong. A small speck of wrong redshi`s is enough to mess up cosmological constraints. z spec 2 1.6 1.2.8.4 R > 5. 1 1-1 1-2 1-3 1-4 Sample used in the plot has 98.6% correct redshi`s and cons2tutes 6% of total sample..4.8 1.2 1.6 2 z true Case study: Simula2ons of DES photometry + VVDS- like spec- z s R: cross- correla2on parameter (measures redshi` confidence) Cunha, Huterer, Lin, Busha, Wechsler et al, in prep.
N(z spec ) For typical exis2ng spectroscopic samples, sample variance is significantly larger than shot noise..1.8.6.4.2 N(z spec ) LSS random.2.4.6.8 1 1.2 1.4 Redshift 1 deg 2 Cunha, Huterer, Busha & Wechsler arxiv: 119:5691 Figure 1. Normalized spectroscopic redshift distribution for the full data. The red (light gray) error bars show the 1-σ variability in the redshift distribution for contiguous 1 deg 2 angular patches. The blue (dark gray) error bars show the variability in the redshift distribution assuming random samples of with the same mean number of objects as the 1 deg 2 patches. We assume that only a 25% random subsample of each patch is targeted for spectroscopy, yielding about 1.2 1 4 galaxies per patch on average.
Spectroscopic simula4ons N- body + photometry + spectra N- body + photometry: BCC sims Used K- correct built- in spectra, added noise, and derived spectroscopic redshi`s using rvsao code.
Survey Calculator Number of patches 1 1 5 1 2 4 8 16 32 64 x 1 3 1/4 deg 2 1/8 deg 2 1/32 deg 2 Magellan VLT σ 95 ( bias ) = 1. 1 1 1 1 Assuming fiducial σ(w)=.35, and perfect spectroscopic selec2on. gals/patch Cunha, Huterer, Busha & Wechsler arxiv: 119:5691
An example: - Template photo- zs. - Calibra2on using one field with 1 deg 2. - Weak Lensing shear- shear tomography. - Difference between true P(z s z p ) and that of calibra2on sample generates biases in cosmology. LSS in one 1deg 2 sample w- bias for fixed ΔP(z s z p )=.1 ΔP(z s z p ) = P(z s z p ) phot - P(z s z p ) train w- bias for ΔP(z s z p ) of Patch 37