Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc...
Transcript of Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc...
![Page 1: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/1.jpg)
Yao-Yuan Mao (Stanford/SLAC ➡ PITT PACC)
Simulations and the Galaxy–Halo Connection
@yaoyuanmao | yymao.github.ioSCMA6 @ CMU 6/10/16
![Page 2: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/2.jpg)
2
My collaborators at Stanford/SLAC
Joe DeRoseBen Lehmann (➡ UCSC)Vincent Su
Marc Williamson (➡ NYU)Hung-I Eric Yang
Risa WechslerPhil Marshall
Matthew Becker
![Page 3: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/3.jpg)
Cosmological Dark Matter Simulations
3
MW-size zoom-in simulations[YYM+ 2015]
Dark Sky Simulations[Skillman+ 2014]
![Page 4: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/4.jpg)
Dark Matter Halos
● High-density regions of dark matter
● Nests of visible galaxies
● Has hierarchical forming/merging processes
● Identified by a “halo finder” (e.g. Rockstar)
![Page 5: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/5.jpg)
(Yes, this actually happens in conferences.) 5
![Page 6: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/6.jpg)
Why do we model the Galaxy–Halo Connection?
➔ Cosmology and large-scale structure- Galaxy clustering- Lensing- Intensity maps
➔ Galaxy formation physics- Feedback, cooling, and many
processes- Dwarf galaxies
6
To marginalize over the GHC
To predict and interpret the GHC
[Springel+ 2006]
![Page 7: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/7.jpg)
Galaxy–Halo Connection
➔ Dark matter-only simulations provide information about halos. How about galaxies?
➔ How do galaxy properties depend on halo properties?
➔ What are the difficulties?
Hydrodynamical simulations [cf. Snyder’s talk today]
➔ computationally expensive(difficult to marginalize over)
➔ include most physics(but still some subgrid models)
Semi-analytical models
➔ build galaxies along the merger histories➔ controlled by “physical” parameters
Empirical models:
- Halo Occupation Distribution- (Subhalo) Abundance Matching
➔ fewer parameters, fits to statistics directly(“data-driven”)
➔ minimalist’s approach7
![Page 8: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/8.jpg)
Example: Connecting galaxies and halos with the Abundance Matching framework
➔ What are the “parameters” in abundance matching?
➔ Can we associate physical meanings to these parameters?
➔ Can we marginalize over these parameters?
➔ Can we do the two simultaneously?
8
To marginalize over the GHC
To predict and interpret the GHC
[cf. Wolfgang’s talk on Thur]
![Page 9: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/9.jpg)
(Subhalo)Abundance Matching
Connects halos and galaxies assuming:
➔ galaxies live in halos and subhalos
➔ one halo property is strongly correlated with one galaxy property
Kravtsov+ 2004Vale & Ostriker 2004, 2006Conroy+ 2006Reddick+ 2013
L M
➔ Matching the halo proxy (e.g. mass) with galaxy luminosity (or stellar mass) at the same number density.
➔ Only the rank of the halo proxy matters.
➔ 1-point function is matched by construction, but other statistics are not
9
![Page 10: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/10.jpg)
Choice of the Matching Proxy
➔ Agreement b/w observed and predicted 2-point correlation functions (or other statistics) provides an evaluation on the proxy.
➔ It was found that “V-peak” or “M-peak” work the best, among others.And “V-peak” is better than “M-peak”.
10[Reddick+ 2013]
![Page 11: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/11.jpg)
Problems with discrete choices
➔ Attaching too much physical meaning to a specific choice
➔ Cannot calculate inference, nor marginalize over the uncertainty associated with proxy choice
11
M-peakV-peak
![Page 12: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/12.jpg)
Interpolating matching proxies
➔ choosing between “mass” or “maximal circular velocity” is changing the concentration dependence
➔ parametrize this freedom by “α”
➔ changing “α” changes how we rank halos
12
[Lehmann, YYM+ 2015]
Concentration dependence
M-peak V-peak
M-peak
V-pe
ak
α
![Page 13: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/13.jpg)
Constraints from large-scale structure
13
➔ Luminosity-selected sample built from SDSS DR7 [Reddick+ 2013]
➔ Fitting projected 2-point correlation function
![Page 14: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/14.jpg)
Constraints on α and scatter
➔ “M-peak” and “V-peak” are both excluded
➔ α > 0 implies that data requires some concentration dependence (assembly bias)
➔ previous studies on AM commonly used 250 Mpc/h boxes
14
Joint constraint on α and scatter from four luminosity-selected samples.
(p = 0.05)
[Lehmann, YYM+ 2015]scatter
M-peak V-peak
![Page 15: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/15.jpg)
Choosing an epoch for the halo proxy
➔ Subhalos are stripped when/after entering the host halo
➔ Stellar content lives in a deeper potential and is less affected
[Behroozi+ 2013] 15Later in time
![Page 16: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/16.jpg)
Interpolating among multiple matching proxies
16
[YYM+ (in prep.)]
➔ β controls the choice b/w using proxy at peak or now.
➔ Constraints from 2-pt correlation function, with a fixed scatter of 0.15 dex.
➔ Allowed region of α increases significantly
➔ β is well constrained to be close to 1 (peak)
0: mass 1: V-max
β
α
0: now 1: peak
![Page 17: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/17.jpg)
What have we learned?
➔Even with abundance matching, which has strong constraints from the framework itself (when compared to HOD, SAM, or hydro sims), there is still much freedom when mapping galaxies to dark matter halos [for HOD, cf. Hahn’s poster on Thur]
➔ Should avoid over-interpreting the physical meaning of a particular choice of matching proxy.
➔ This “freedom” in the galaxy-halo connection, when constrained by data, provides insights into galaxy formation physics:● do galaxy properties depend on halo formation history (at fixed mass)?● does initial stripping of subhalo strongly affect the galaxy within?
➔ The “freedom” should be also marginalized for other inferences.
17
![Page 18: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/18.jpg)
Challenges for simulations
18
➔ Finite volume of simulations results in stochastic errors (sample variance)
➔ Finite resolution of simulations results in systematic errors (bias)
[van den Bosch+ 2016]
[Lehmann, YYM+ 2015]
![Page 19: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/19.jpg)
Trade-off in cosmological simulations
The computational cost of a dark matter-only simulation is ~ N log N
Given limited resources, it is always a trade-off between:● a higher resolution● a larger volume
Resolution
Volume
![Page 20: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/20.jpg)
Resolution requirements
➔ To construct a volume-limited samples from a magnitude-limited survey, the dimmer galaxies we look at, the smaller the total volume is.
➔ To model dimmer galaxies, we need to resolve smaller halos
20[YYM, Yang+ (in prep.)]
Larger volume Dimmer galaxies
![Page 21: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/21.jpg)
Resolution requirements
➔ The sample variance (stochastic error) increases as the volume decreases
➔ Hence, the most informative sample is not the dimmest sample
➔ Simulations should balance b/w volume and resolution
21[YYM, Yang+ (in prep.)]
larger volume(less stochastic error)
dimmer galaxies(more systematic error)
![Page 22: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/22.jpg)
Despite the challenges, there are still other ways to tackle the galaxy–halo connection for specific questions
➔ Can we infer other galaxy properties if its luminosity is already known?Color? Morphology?
➔ Can we infer halo properties for specific galaxies from their observed properties?
22
![Page 23: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/23.jpg)
Inferring galaxy color from large-scale density➔ Galaxy color and its environment are correlated [cf. Chen’s talk on Wed]
➔ Can we find informative “features” that are predictive for galaxy color?
➔ Can we find “features” that are less sensitive to resolution limits?
23
[Su, DeRose, YYM+ (in prep.)]
![Page 24: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/24.jpg)
Predicting clustering signals for galaxies of different colors (sSFR)
24
[Su, DeRose, YYM+ (in prep.)]
![Page 25: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/25.jpg)
Inferring halo properties from kinematics➔ The kinematics of our neighbor galaxies provide information on the mass
profile of the dark matter halo [cf. Eadie’s talk today]
➔ How to construct a proper prior? Can simulations help?
25[Williamson, YYM+ (in prep.)]
![Page 26: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/26.jpg)
Posterior distributions of halo mass (left) and concentration (right)Using LMC and SMC kinematics and maximal circular velocities(update to Busha+ 2011)
26halo mass concentration
[Williamson, YYM+ (in prep.)]
![Page 27: Simulations and the Galaxy–Halo Connection - CMU …cschafer/SCMA6/Mao.pdf · · 2016-06-10Marc Williamson ( NYU) Hung-I Eric Yang Risa Wechsler Phil Marshall ... Why do we model](https://reader033.fdocuments.in/reader033/viewer/2022052917/5aa802a77f8b9ab1258b4807/html5/thumbnails/27.jpg)
Summary
➔ Dark matter simulations provide robust and scalable predictions of the matter distribution for a given cosmology
➔ Need to model or to marginalize over the uncertainties in the galaxy–halo connection
➔ Empirical models such as abundance matching provide a framework for the above (but we’re not doing it yet)
➔ Trade-off b/w resolution and volume is always challenging. Need to balance different sources of error
➔ Other statistical/ML methods can be applied to simulations to probe the galaxy–halo connection
and thank you!
Yao-Yuan Mao (Stanford ➡ PITT)@yaoyuanmao | yymao.github.io