What the notebooks depend on

In your browser

Nothing you have to install. Each notebook bundles:

Component	Version	Role
marimo	0.23.9 (pinned)	reactive notebook runtime and WASM export
Pyodide	0.27.7	CPython 3.12 compiled to WebAssembly
NumPy, SciPy	as shipped by that Pyodide build	numerics
Plotly	as shipped by that Pyodide build	interactive figures, including rotatable 3-D attractors
`chaoslib`	0.1.0	this book’s own shared numerics, installed from a bundled wheel

The marimo pin is the load-bearing one. marimo’s version determines which Pyodide build every export bundles, and that Pyodide ships its own NumPy and SciPy. So pinning marimo pins your browser environment, not just the authors’ — which is what makes a figure in this book reproducible. The other runtime versions are deliberately left to float, because requesting a specific NumPy that the bundled Pyodide does not have would break the notebook with no error visible at build time.

`chaoslib`

All the numerics live in one small library rather than being repeated across notebooks, so that the equation printed in a chapter is provably the equation being stepped, and one test suite covers all of it.

Module	Provides
`systems`	right-hand sides and Jacobians: Lorenz 63 and 96, pendulum and double pendulum, logistic and Hénon maps
`integrate`	vectorised fixed-step RK4 for ensembles; adaptive integration for single trajectories
`lyapunov`	Lyapunov spectra, finite-time exponents, doubling times, Kaplan–Yorke dimension, KS entropy
`errorgrowth`	saturation and the logistic error-growth model
`dimension`	correlation dimension from a sampled trajectory
`information`	entropy, relative entropy, mutual information
`adjoint`	tangent linear and adjoint propagators; singular vectors
`assimilate`	Kalman filter, 3D-Var, 4D-Var, EnKF, localisation
`ensemble`	ensemble construction, spread–skill, rank histograms, CRPS, Brier score
`plotting`	the book’s semantic colour palette and figure styling

The library carries 90 tests, anchored wherever possible to identities that hold exactly rather than approximately — the Lyapunov exponents summing to the trace of the Jacobian, the adjoint identity, energy conservation, $\det\mathbf{M} = e^{\tau\,\mathrm{tr}\mathbf{J}}$ — and otherwise to published values with the source named. Two real bugs were caught this way during construction, and both would have produced plausible-looking figures.

If you want to run it locally

git clone https://github.com/anichaospred/anichaospred.github.io
cd anichaospred.github.io/chaos-book
pip install -r requirements.txt
make test        # the chaoslib suite
marimo edit notebooks/ch06_lorenz63.py

requirements.txt pins the authoring environment exactly — that governs the test suite and local editing, not what readers receive.