JaxPM_highres/design.md

53 lines
1.4 KiB
Markdown
Raw Permalink Normal View History

2021-07-16 23:15:27 +02:00
# Design Document for JaxPM
2021-07-17 00:59:03 +02:00
This document aims to detail some of the API, implementation choices, and internal mechanism.
2021-07-16 23:15:27 +02:00
2021-07-17 00:59:03 +02:00
## Objective
2024-07-09 14:54:34 -04:00
Provide a user-friendly framework for distributed Particle-Mesh N-body simulations.
2021-07-17 00:59:03 +02:00
## Related Work
This project would be the latest iteration of a number of past libraries that have provided differentiable N-body models.
- [FlowPM](https://github.com/DifferentiableUniverseInitiative/flowpm): TensorFlow
2024-07-09 14:54:34 -04:00
- [vmad FastPM](https://github.com/rainwoodman/vmad/blob/master/vmad/lib/fastpm.py): VMAD
2021-07-17 00:59:03 +02:00
- Borg
In addition, a number of fast N-body simulation projets exist out there:
- [FastPM](https://github.com/fastpm/fastpm)
- ...
## Design Overview
### Coding principles
Following recent trends and JAX philosophy, the library should have a functional programming type of interface.
### Illustration of API
Here is a potential illustration of what the user interface could be for the simulation code:
```python
import jaxpm as jpm
import jax_cosmo as jc
# Instantiate differentiable cosmology object
cosmo = jc.Planck()
# Creates initial conditions
inital_conditions = jpm.generate_ic(cosmo, boxsize, nmesh, dtype='float32')
# Create a particular solver
solver = jpm.solvers.fastpm(cosmo, B=1)
# Initialize and run the simulation
state = solver.init(initial_conditions)
state = solver.nbody(state)
# Painting the results
density = jpm.zeros(boxsize, nmesh)
density = jpm.paint(density, state.positions)
```