Skip to content

Software, Notebooks, and Benchmarks

Software and notebooks become trustworthy only when the calculation is reproducible, benchmarked, and connected to the canonical physics page it supports. A notebook that produces a plausible plot but does not state its Hamiltonian, units, method, environment, and validation check is not yet part of the reference layer.

This section connects analytic pages, computational notebooks, package choices, and benchmark evidence.

A computational artifact belongs in this section only when it identifies:

  • the canonical page or formula being reproduced,
  • the numerical method and approximation,
  • the unit and convention choices,
  • the input parameters and random seeds when relevant,
  • the exact environment or reproducible dependency recipe,
  • the benchmark target and pass criterion,
  • the known limitations and failure modes.

Artifacts that are useful but not yet reproducible can be listed as planned or conceptual. They should not be cited as evidence.

The first committed notebook family is notebooks/wave-mechanics-canonical-systems/. Those notebooks support the Wave Mechanics and Model Systems volume and are summarized by the volume-specific Numerical Notebooks Index.

The many-body family is not yet committed. Its Reproducible Notebooks page records planned paths and promotion requirements without presenting them as available artifacts.

The Computational AMO and Quantum Chemistry chapter now supplies the domain-specific benchmark ladder for atomic structure, molecular electronic structure, quantum-optical dynamics, and AMO-platform simulations. Its notebook pages remain planned until executable artifacts satisfy the admission rule here.

The Reference pages here generalize that contract so future notebooks from approximation theory, quantum information, AMO physics, many-body physics, and open systems use the same evidence discipline.

  • A plot is exported without the notebook that generated it.
  • A notebook passes once but records no environment or package versions.
  • A benchmark has no predeclared tolerance.
  • A package-specific convention is silently imported into a canonical formula page.
  • A stochastic result omits the seed, sample size, uncertainty estimate, or autocorrelation check.
  • A computational page cites a package paper but not the versioned package documentation used.