search

duncandc / intrinsic_alignments

project to measure and model the intrinsic alignments of galaxies in both observations and simulations
2 stars 1 forks source link

allow for host halo ellipticity--alignment strength dependence for central alignments. #13

Open duncandc opened 6 years ago

duncandc commented 6 years ago

I think @jablazek is particularly interested in this idea.

general trend should be that less elliptical haloes display less alignment with central galaxies.

aphearin commented 6 years ago

One disadvantage of a Monte Carlo based on sample-and-rejection is that it's harder to implement residual correlations. The natural CAM way to implement this would be to use sliding_conditional_percentile to calculate Prob(< e | Mvir), and use those as the values passed to the CDF inverse function. How were you thinking of implementing such correlations?

duncandc commented 6 years ago

1.) Some new code would need to be written and it might be slower, but I don't see why this couldn't be done. 2.) I don't actually think the CAM framework is right for this. The misalignment distribution is not observable, so a forward model approach where alignment_strength = f(m_vir, e_halo) seems reasonable to me.

aphearin commented 6 years ago

Well, it just depends on how you want to get the nonlinear correlations in there. You can still properly forward-model from CAM, so long as you Monte Carlo draw from modeled/parameterized distributions instead of directly from the data.

Your approach to have the PDF parameters themselves depend on {e, M} should also work; I was just asking what the strategy is to minimize the number of parameters.

duncandc commented 6 years ago

I rescind my previous comment point (2). A CAM like approach is probably ideal for this issue. we really don't want to add too many parameters if we ever want to MCMC constrain this model.