vbjax-dynamics 0.1.0

JAX-based integrators for ODEs, DDEs, SDEs, and SDDEs

vbjax_dynamics

A JAX-based library for numerical integration of dynamical systems.

Note: This package contains code adapted from vbjax by INS-AMU. The core integration functions in loops.py are derived from vbjax's implementation.

Features

• ODE Integration: Ordinary Differential Equations

  • Efficient loop-based integrators with JIT compilation
  • Full support for jax.vmap for parallel trajectory computation

• SDE Integration: Stochastic Differential Equations

  • make_sde(): Integration with pre-generated noise arrays
  • make_sde_auto(): Automatic noise generation from random keys
  • Euler-Maruyama scheme
  • Fully reproducible with random seeds

• DDE Integration: Delay Differential Equations

  • Support for fixed delays
  • History function interpolation

• SDDE Integration: Stochastic Delay Differential Equations

  • Combined stochastic and delay dynamics

• Continuation Methods: Parameter continuation for bifurcation analysis

• Configuration Utilities: Easy control over JAX settings

  • configure_jax(): Global configuration
  • precision_context(): Temporary precision changes
  • print_jax_config(): Diagnostic information

• JAX-Native:

  • JIT compilation for speed
  • Automatic differentiation ready
  • GPU/TPU compatible
  • Pure functional approach

Installation

pip install vbjax_dynamics

For development:

pip install -e ".[dev]"

Quick Start

import jax.numpy as jnp
from jax import random, vmap
from vbjax_dynamics.loops import make_sde_auto

# Define Ornstein-Uhlenbeck process
def drift(x, p):
    return -p[0] * x  # -theta * x

def diffusion(x, p):
    return p[1]  # sigma

# Create integrator
dt = 0.01
step, loop = make_sde_auto(dt, drift, diffusion)

# Single trajectory
x0 = 2.0
params = (1.0, 0.5)  # (theta, sigma)
n_steps = 1000
key = random.PRNGKey(42)

trajectory = loop(x0, n_steps, params, key)
print(f"Final value: {trajectory[-1]:.4f}")

# Multiple trajectories in parallel with vmap
n_traj = 100
keys = random.split(key, n_traj)
trajectories = vmap(lambda k: loop(x0, n_steps, params, k))(keys)
print(f"Mean: {jnp.mean(trajectories[:, -1]):.4f}")

For more examples, see the examples/ directory.

Documentation

• Examples and Tutorials: Complete guide with detailed examples for ODE, SDE, DDE, and SDDE integration
• Testing Guide: Information about running and writing tests
• Acknowledgments: Credits and attribution

Acknowledgments

This package includes code adapted from vbjax, developed by the Institut de Neurosciences de la Timone (INS-AMU). We are grateful for their work on efficient JAX-based numerical integrators.

License

MIT License