Presentation (Thursday, Oct 30): in http://www.iap.fr/users/gam/GAIACHALLENGE/Mamon.pdf

MAMPOSSt figures in http://www.iap.fr/users/gam/GAIACHALLENGE/mamposst_figures.tar.gz (9 MB)

PlumCoreIso_1000_1_easy PlumCoreIso_1000_1_hard PlumCoreIso_1000_1_vhard PlumCoreIso_1000_2_easy PlumCoreIso_1000_2_hard PlumCoreIso_1000_2_vhard PlumCoreIso_1000_3_easy PlumCoreIso_1000_3_hard PlumCoreIso_1000_3_vhard PlumCoreIso_100_1_easy PlumCoreIso_100_1_hard PlumCoreIso_100_1_vhard PlumCoreIso_100_2_easy PlumCoreIso_100_2_hard PlumCoreIso_100_2_vhard PlumCoreIso_100_3_easy PlumCoreIso_100_3_hard PlumCoreIso_100_3_vhard PlumCoreIso_100_4_easy PlumCoreIso_100_4_hard PlumCoreIso_100_4_vhard PlumCoreIso_100_5_easy PlumCoreIso_100_5_hard PlumCoreIso_100_5_vhard PlumCoreIso_100_6_easy PlumCoreIso_100_6_hard PlumCoreIso_100_6_vhard PlumCoreIso_100_7_easy PlumCoreIso_100_7_hard PlumCoreIso_100_7_vhard PlumCoreIso_100_8_easy PlumCoreIso_100_8_hard PlumCoreIso_100_8_vhard PlumCoreIso_100_9_easy PlumCoreIso_100_9_hard PlumCoreIso_100_9_vhard PlumCoreIso_100_10_easy PlumCoreIso_100_10_hard PlumCoreIso_100_10_vhard PlumCoreOM_1000_1_easy PlumCoreOM_1000_1_hard PlumCoreOM_1000_1_vhard PlumCoreOM_1000_2_easy PlumCoreOM_1000_2_hard PlumCoreOM_1000_2_vhard PlumCoreOM_1000_3_easy PlumCoreOM_1000_3_hard PlumCoreOM_1000_3_vhard PlumCoreOM_100_1_easy PlumCoreOM_100_1_hard PlumCoreOM_100_1_vhard PlumCoreOM_100_2_easy PlumCoreOM_100_2_hard PlumCoreOM_100_2_vhard PlumCoreOM_100_3_easy PlumCoreOM_100_3_hard PlumCoreOM_100_3_vhard PlumCoreOM_100_4_easy PlumCoreOM_100_4_hard PlumCoreOM_100_4_vhard PlumCoreOM_100_5_easy PlumCoreOM_100_5_hard PlumCoreOM_100_5_vhard

These models are a development of http://arxiv.org/abs/1303.6099. The marginal likelihood $p(D|\Phi)$ is calculated by representing the DF by an arbitrary number of Gaussians of arbitrary weight/location/covariance in action space and marginalising the parameters describing the Gaussians. So, no assumption is made about the light profile (except that it is axisymmetric) or the orbit anisotropy. The implementation is for now still limited to error-free data though.

The results below assume the correct double power-law form for potential with scale radius held at correct value. Unknown parameters are then the inner slope $\gamma$ and scale densty $\rho_0$.

These use the standard “extended Schwarzschild” approach, but using big blocks of orbits in $(E,L^2)$ (or equiv actions) instead of single orbits. The models are fed $(X,Y,V_X,V_Y)$ from PlumCuspIso with 2% errors in $(V_X,V_Y)$ – and infinite uncertainty in $(Z,V_Z)$

Noise in result is probably due to finite resolution of block DF, plus the deficiency of the maximum-likelihood approach used by such “extended-Schwarzschild” methods. And there might be some implementation errors too.