Inferring the Galac/c gravita/onal poten/al with chemodynamics Robyn Sanderson NSF Postdoctoral Fellow Columbia University/NYU The accreted stellar halo is clumpy in ac/on space View in Galac/c coordinates 6 Jr �kpc2 Myr�1� 5 View in ac)on space (using input poten)al) n��497198 M�2.7x1012 M� b�8 kpc 4 3 2 1 0 R.E. Sanderson • 21 July 2015 �5 Lz �kpc Myr � 0 2 �1 5 Review of ac/on-­‐angle variables R.E. Sanderson • 21 July 2015 Ac/ons are most clustered in the correct poten/al � Potential parameters � 1� GM − L + L2 + 4GM b Observations Jr = √ 2 −2E Both Mtrue=2.7x1012 M⊙ R.E. Sanderson • 21 July 2015 The Kullback-­‐Leibler divergence measures clustering DKL (p||q) = Clumpier distribution = larger relative entropy DKL = 1.1 R.E. Sanderson • 21 July 2015 � ln � p(x) q(x) � p(x)dx Smoother distribution = smaller relative entropy DKL = 0.4 Errors blur, but do not destroy the informa/on Without errors With Gaia (pre-launch) errors Sanderson, Helmi, & Hogg, 2015 R.E. Sanderson • 21 July 2015 Results of tests with a mock isochrone halo Sanderson, Helmi, & Hogg, 2015 Log10 conditional probability Without errors With pre-­‐launch Gaia errors 67, 95, 99% confidence equivalents dGC<20 kpc R.E. Sanderson • 21 July 2015 Simulated stellar halo in cosmological poten/al 15 thin streams chosen by hand from tagged Aquarius halo Aq-­‐A (Cooper+2010, Helmi+2011) Sanderson, Hartke, & Helmi, in prep R.E. Sanderson • 21 July 2015 Best-­‐fit halo matches present-­‐day mass profile Model: smooth, spherical NFW *Membership info not used in fit* Sanderson, Hartke, & Helmi, in prep R.E. Sanderson • 21 July 2015 Synergy in observa/ons RVs • Northern Hemisphere: WEAVE on William Herschel Telescope (INT/IAC, La Palma, Canary Islands); DESI at KPNO • Southern Hemisphere: 4MOST on VISTA telescope in Chile (ESO, Cerro Paranal, Chile) • 4m telescope with mul)ple-­‐fiber instrument • ~10 halo stars (per survey) • RV errors <2 km s to V=20-­‐22 • Adds RVs to stars with Gaia proper mo)ons 20% photometric distances • RR Lyrae • Distances to 2%, RV errors 5-­‐10 km s 6 -­‐1 -­‐1 R.E. Sanderson • 21 July 2015 Improved distances increase informa/on, but decrease sampling KIII giants, Gaia parallaxes R.E. Sanderson • 21 July 2015 RRLe distances Observa/ons with larger average dGC do beYer Sanderson, in prep R.E. Sanderson • 21 July 2015 R.E. Sanderson • 21 July 2015 Chemical dimensions Bullock & Johnston Halo 2 + Galaxia (S. Sharma), single stellar popula9ons -­‐> M Giants (mag & color selected) error models: Gaia + RVs + some kind of abundance error (wild guesses) cut error-­‐convolved sample in es9mated energy (~40,000 stars) view here is without observa9onal errors for stars selected as above bad poten9al model: spherical NFW for bulge/disk/halo system! R.E. Sanderson • 21 July 2015 Chemical dimensions Same as before, but now with observa9onal errors error models: Gaia parallaxes/pms, 2 km/s RVs, abundances to 0.1 dex bad poten9al model: spherical NFW for bulge/disk/halo system! R.E. Sanderson • 21 July 2015 Chemical dimensions Same as before, but now with observa9onal errors error models: Gaia parallaxes/pms, 2 km/s RVs, abundances to 0.05 dex bad poten9al model: spherical NFW for bulge/disk/halo system! R.E. Sanderson • 21 July 2015 Chemical dimensions Same as before, but now with observa9onal errors error models: Gaia parallaxes/pms, 2 km/s RVs, abundances to 0.01 dex bad poten9al model: spherical NFW for bulge/disk/halo system! R.E. Sanderson • 21 July 2015 Phase space only (3 dimensions) Without obsv. errors R.E. Sanderson • 21 July 2015 With obsv. errors Adding [Fe/H] (4 dimensions) Without obsv. errors R.E. Sanderson • 21 July 2015 With obsv. errors ([α/Fe] to 0.1 dex) Adding [Fe/H] and [α/Fe] (5 dimensions) Without obsv. errors R.E. Sanderson • 21 July 2015 With obsv. errors ([α/Fe] to 0.1 dex) Adding [Fe/H] and [α/Fe] (5 dimensions) Without obsv. errors R.E. Sanderson • 21 July 2015 With obsv. errors ([α/Fe] to 0.05 dex) Adding [Fe/H] and [α/Fe] (5 dimensions) Without obsv. errors R.E. Sanderson • 21 July 2015 With obsv. errors ([α/Fe] to 0.01 dex) Finding streams in ac/on space Using EnLink (Sharma & Johnston 2009) 1 group found 512/41,028 stars classified did not fine-­‐tune parameters of group finder R.E. Sanderson • 21 July 2015 Finding streams in 5D chemodynamical space abundances to 0.05 dex 5 groups found 1,120/41,028 stars classified R.E. Sanderson • 21 July 2015 Finding streams in chemodynamical space abundances to 0.01 dex (yeah) 10 groups found 5,427/41,028 stars classified R.E. Sanderson • 21 July 2015 Finding streams in chemodynamical space R.E. Sanderson • 21 July 2015 For discussion • BeIer choice of stellar popula9on? • BeYer poten/al model? • Other things I am doing wrong with abundances? R.E. Sanderson • 21 July 2015 Advantages • No need to assign membership of stars in streams (membership probability is an output) • Robust to adiaba/c /me-­‐evolu/on of host, substructure, some chao/c diffusion • Can use any informa/ve parameter (i.e. abundances) Disadvantages • Needs 6D phase space info (plus whatever else) • Not a forward model (no way to treat uncertain/es) • Derailed by large (>50%) smooth component • Limited to integrable poten/als R.E. Sanderson • 21 July 2015 The takeaway • Tidal streams are sensi)ve to the poten)al of the Galac)c dark mager halo • Streams form clumps in ac)on space that s)ll look clumpy at Gaia precision (δπ/π < 0.2) • Maximizing D KL of ac)on distribu)on iden9fies the poten9al parameters • Rela)on between D KL and posterior probability lets us set confidence intervals • Expected number of MW streams is enough • Stream membership info not required R.E. Sanderson • 21 July 2015