> # KITP darkmatter24: Combining Probes
>

---

## **Pre-discussion**

* Probes of dark matter (DM): dwarf galaxy population, strong lensing, streams, lyman alpha, UVLF, line intensity mapping, weak lensing, galaxy clustering, CMB (most are aiming at reconstruction of P(k). See Fig. 1. - placeholder)
* Ideal goal: combining simulations to span a range of scales and redshifts. Some parts of this task are more readily achievable than others.
* There is a distinction between spanning scales and spanning redshifts, in terms of modeling. They present different challenges.
* Forward models can be developed independently by different groups, for different observational probes, ideally in a modular manner so they can be swapped out in analyses. Is this possible at present?
* Three distinct “buckets” of tasks for combining observables:
* 1) small scales & low z
* 2) small scales & large z-span (from dwarfs to 21-cm)
* 3) high-z & large scales

* Potential questions for discussion:
1. Do we understand CV on small scales?
2. What modeling approaches might mitigate CV and selection function-related concerns?
3. Concerns re: errorbars and models, for dwarfs and strong lensing?
4. Should we quote 1 / 3 / 5 sigma uncertainties on all constraints?
5. How can the community encourage reproducible science in the DM field?
6. What shared code/data system could we emulate, and which parts of small-scale modeling pipelines are modular ?
7. Can we build consistent models for combining different z/k probes? Eg multi-scale zoom in sims, or SAMs, common parameters?
8. Minimal non-trivial examples for building joint likelihood?
9. Practicalities?


---

## **Part 1 (Jun 13): Overview of key questions**

- proposal to build a toy model of a non-trivial data combination at the workshop.
- proposal to place all astro nuisance parameters on the same plot. is this possible?
- should we worry about “cosmic variance”, aka variations between individual systems we observe, based on, e.g. assembly history or large-scale environments, or other hidden variables? on the observational side, we already have 100 MW systems, in SAGA.
- on the modeling side, there is already a body of work showing how lack of knowledge of the host halo mass is a key uncertainty, for example.
- is there a fundamental limit to how many parameters are sufficient to model observables well (eg dwarf population)? there is a mix of optimism and pessimism about this.
- “when do we get in trouble simulating the milky way?” it’s hard to answer this generically, depends on the observable in question. for example, if you think about the special anisotropy of satellites, LMC matters a lot.
- stellar streams might be also on the more sensitive side; but so are satellite abundances; so, which observable is not sensitive? we didn’t discuss concrete examples here.
- point was brought up that the angles on which satellites are brought into the MW halo matter a lot; their orbits will affect their evolution etc.
- there are ways to formally quantify this using simulations; example of velocity fields in the local groups is given.
- problems come from: too many nuisance params - and many are very correlated. no consensus yet on which ones are the most important ones, but people are generally hopeful that there is a way to figure this out. but this is exactly why we may ultimately rely on a combination of probes when talking about dark matter physics discovery.
- it was brought up that there is lots of motivation to show that we can do inflationary cosmology and other fundamental physics that cosmology has traditionally tackled - but with small scales.
- breaking degeneracies should prob be the main reason why we combine the probes.
- it was suggested that we don’t worry about the cross-k covariances for right now.
- except maybe covariance between satellites and streams.
- discussion goes back to the question of whether we can have the same params for across redshifts? are there examples? reionization was discussed at some length. high z lya + dwarfs, can they all be combined?
- lots of points brought up about local vs global reionization (shown to be fairly local in the past but this may be worth revisiting - this is encouraging).
- the range of reionization z’s in mw seems to be comparable from MW to MW. question was asked: are we reionized by virgo or not? this may be possible to figure out soon.
- but what about other parameters, like feedback?
- understanding escape fractions is important but possible (from halos themselves, SFR can depend on that).
- how much should we be thinking about getting things right?
- CMB: analysis has a lot of subtlety; perhaps not even the CMB is beyond these same issues.
- a key question: how do we precisely understand what the noise is, in order to understand how large a signal we can detect, with any of the probes.
- suggestion to create mock observations - or a data challenge?
- modularity - is also suggested
- how many people is usually on the pipeline building tasks in large collaborations? just a few key contributors. we’re not people-power limited.
- some things have common systematics; but many probes do not. especially if at different z’s.
- it is possible that we need to simulate subhalos and streams in the same framework.
- PGM (probabilistic graphical model) would be good to create to better understand relationships between variables and observables and between different observables.
- question was asked about how much do different zoom ins share subgrid physics? some but not all and perhaps not at a level that’s satisfying, at the moment.
- what will discovery look like? should we create a data challenge to recover a signal in multiple probes?
- some people have the sense that multiple probes have a large discovery potential and any one of them sees an anomaly, that will provide lots of motivation for others to follow up on it… does this mean we don’t need to do anything to prepare for this?
- another related question: what would a detection with lyman alpha forest look like? what should we worry about?
- we already know that lya data from eboss are inconsistent with LCDM; but new work shows that modeling the 1D spectrum actually recovers better consistency; so, we’re already in teh regime when these systematic matter. the sense is that the community is now turning towards a more careful modeling of the lya signal and understanding potential systematics. but there are dark matter models that will affect lya but NOT affect satellites - so we might need to be digging deeper into the data in order to test whether an anomaly is a systematic or new physics, in that case.
- there may be multiple probes that probe at the same level, but it is unclear when this might be the case.
- some people have the sense that we were looking so far for strong signals, and now we are trying to push for precision near-field cosmology. and this is a new regime.
- example of strong lensing + satellite joint analysis: amplitude of subhalo mass functions and xx were the only parameters where both probes had something meaningful to say, so they marginalized over other parameters, and then multiplied pre-marginalized likelihoods. it is *much* harder to unpack and consistently model both likelihoods and data sets.
- what are the next params that such joint analysis might care about? probably something characterizing the stripping…
- example was brought up of attempted combination of lensing + clustering + satellite kinematics, but relevant observables were completely off. is this a cautionary tale, or something we want to try to adress, in the future?
- shared code and simulation base is a critical starting point. general agreement on this?
- a point was brought up about how models that can be used for inference (HOD, gal-halo connection) are a good empirically-driven starting point. there is agreement on the fact that they enable inference to be done with small scale data.
- mixed feelings on whether empirical models are sufficiently flexible to allow for precision science and lead to discovery. suggestion to develop empirical models for other observable, perhaps strong lensing.
- hydro sims would be used to root the empirical models.
- empirical approaches - but also their refinement may play a critical role in future inference.
- one fact about this diverse community: there aren’t many people who understand this many fields all at once.
- *Action item: What does each observable require in terms of modeling (a list of modeling dependencies/variables and potential sources of systematics)? To be continued tomorrow.*

---
## **Part 2 (Jun 14): Hack on PGM of observables**

* **Dwarfs**
1. Observables:
* galaxy luminosity function
* radial distribution
* size distribution
* velocity dispersion
* metallicity
* starformation history
2. Physics:
* gas cooling
* reionization
* feedback
* IMF
* tidal evolution
* dynamical friction
* merger history
* disk
* magnetic fields

* **Strong lensing**
1. Observables
* Convergence/shear map?
* Flux ratios
2. Observational systematics
* lens properties (Mh, environment [central, satellite], stellar mass, formation history [profile, satellites, SF history])
* source properties (morphology, clumpiness)
3. Theoretical systematics
* subhalos (infall dn/dm, Vmax, distruption/stripping, internal feedback, host-subhalo interactions)
* LOS halos (dn/dm, $\xi(r)$, $V_\text{max}$, $M_*-M_h$)

* **Cosmic shear**

* intrinsic alignments
* modulation of $P_\text{NL}$
* (dwarf) galaxy lensing
* $\Delta \Sigma(r)$
* lens sample selection function
* total density profile ($V_\text{max}(r_\text{max})$, $f_\text{sat}$, dn/dm, galaxy luminosity function, $M_*-M_h$)

* **Streams**
1. Modeling
* similar to satellites (but at different mass scale and formation history): infall, dwarf stripping, disruption, ...
* orbits
* progenitors
* other perturbers

2. Observables
* number and distribution
* velocity distribution
* phase space
* morphology
* gaps
* metallicity

3. systematics
* bar
* giant molecular clouds
* dwarfs
* globular clusters
* black holes
* MW formation history (LMC orbit, infall)

* **CMB**
* dust
* synchrotron
* atmospheric variability
* astrophysics with SZ cluster
* galactic foreground modeling
* CIB
* extragalactic sources

* **High-z galaxies**
1. Observables
* UVLF
* UV slopes
* 2pt function (clustering)
2. Modeling
* feedback
* IMF
* SF history
* dust
* HMF
* halo growth (within a given cosmology)
* galaxy-halo connection
* mergers

* **21-cm**
1. Observable
* global signal
* power spectrum
* distrinction between EoR and cosmic dawn
2. Modeling
* HMF
* SF history
* IMF
* feedback
* X-ray heating
* escape fraction
* radiative transfer
* molecular cooling

* **dwarf velocity dispersions**
1. Observable
* high res stellar spectra
2. Systematics
* number of member stars
* membership of stars to the halo?
* stellar binaries?
* modeling phase space density in UFDs?
* reconstructing underlying mass distribution?