search

HITS-AIN / Spherinator

https://space.h-its.org
Apache License 2.0
3 stars 3 forks source link

Prototype use cases description #30

Closed sebastian-tg closed 7 months ago

sebastian-tg commented 8 months ago

Description of possible use cases for Spherinator

  1. Understanding the effect of large-scale environment on DM halo structure and galaxy properties

    1. Train Spherinator to reproduce 3D structure of DM in simulated galaxies
    2. Select minimum number of output features based on minimum required accuracy of reconstruction (i.e. 5-10%)
    3. Paint haloes in feature space using traditional interpretable halo properties (mass, concentration, spin, etc) to help interpret dominant features
    4. Color points by large scale environment to find effect and produce predictions
    5. Do the same for 3D baryon distribution (i.e. stars or gas)

  2. Galaxy simulation parameter search: produce a robust metric for simulations to understand how close a given sub grid model gets to reproduce the properties of galaxies

    1. Use Spherinator to represent stellar distributions of sims with minimum dimensionality (i.e. 2D or 3D)
    2. Compare distribution of points of simulations and observations until a model is found that is statistically consistent
    3. Model can then be used to deconstruct the physics of galaxy formation and to reconstruct histories of observed galaxies
    4. Can also generate mocks on top of N-body sims

  3. Detect outliers in simulations to test the input physics at the extremes of the distributions

  4. Find compact representation of DM halo structure to provide robust models for Bayesian tests of CDM predictions regarding the haloes of observed galaxies

    1. Use Spherinator to find minimum representation of entire sample of haloes in simulations with specific physics
    2. Encoder network is then able to generate distributions that can be compared to mass models of galaxy kinematic data
      1. Is the data consistent with the simulated distribution?
      2. Eliminates the need to collapse diversity into mass, concentration axes
      3. Allows to obtain the closest match of an observed RC in a simulation for consistency tests
BerndDoser commented 7 months ago

Instead of using GitHub issues, I recommend using a markup file in the space-dev repository or Overleaf directly for collaborating on text.