Skip to content

gregreen/deep-potential

master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 

deep-potential

Deep learning for gravitational potentials, based on a snapshot of well-mixed tracer particles in phase space.

The basic idea of this approach is to first model the distribution function of the tracers using a normalizing flow. One can then calculate gradients of the distribution function at a large number of points in phase space. Then, we find the potential that renders the distribution function stationary at these points. We model the potential using a feed-forward neural network, which is both extremely flexible and easily differentiable. This latter property is critical, as the collisionless Boltzmann equation contains gradients of the potential (and of the distribution function).

See notebooks/plummer_sphere_example.ipynb for an explanation of the method and a demonstration with a simple toy system - the Plummer Sphere with isotropic velocities.

This version is implemented in Tensorflow 2.x. There is a matching PyTorch implementation at tingyuansen/deep-potential.

About

Deep learning for gravitational potentials, based on well-mixed tracers in phase space.

Resources

License

Stars

Watchers

Forks

Packages

No packages published