Analysis pipeline

lyapower turns the Lyman-alpha flux field of a Nyx simulation into a fitted 3D flux power spectrum and its Arinyo-i-Prats et al. 2015 model parameters.

flowchart TD
    A["Nyx simulation<br/>(optionally rebinned, utils_fitter)"] --> B["gimlet<br/>P3D(k, mu) / P1D / P(k)"]
    B --> C["power_spectra<br/>read gimlet / genpk / ascii"]
    C --> D{"multiple<br/>resolutions?"}
    D -- yes --> E["splice_1D / splice_3D<br/>combine boxes"]
    D -- no --> F["rebin / cut"]
    E --> F
    G["CLASS / cosmoprimo<br/>linear matter P(k)"] --> H["fitter<br/>Pf_model = Pm x bias x Arinyo(D0/D1)"]
    F --> H
    H --> I["iminuit fit<br/>b, beta, non-linear params"]

Stages

Stage	Module	What it does
(Pre) rebinning	utils_fitter	Optionally downsample a Nyx/HDF5 field (e.g. optical depth → transmitted flux) slice by slice before measuring the power.
Read	power_spectra	Load gimlet / genpk / ascii files into MatterPowerSpectrum / FluxPowerSpectrum objects (P(k) or P(k, μ)).
Splice	power_spectra	Combine low- and high-resolution boxes into a single spectrum (splice_1D, splice_3D) to extend the dynamic range.
Rebin / cut	power_spectra	Rebin in k (or k, μ) and restrict the fitted range (rebin_arrays, rebin_2d_arrays, cut_extremum).
Linear P(k)	CLASS	Compute the linear matter power spectrum with CLASS (MyClass) or cosmoprimo (CosmoprimoInterface), with a no-wiggle option for BAO.
Model + fit	fitter	Build the flux model Pf_model (linear Pm × bias/RSD × Arinyo non-linear D0/D1) and fit b, β and the non-linear parameters with iminuit.

The Arinyo model

The measured flux power spectrum is modelled as the linear matter power spectrum times a bias/RSD factor and a non-linear correction:

• D0 — the Arinyo non-linear term parameterised by (k_nl, a_nl, k_p, a_p, k_v0, a_v0, k_v1, a_v1);
• D1 — an alternative parameterisation (q_1, q_2, k_v, a_v, b_v, k_p);
• BAO variants add broadening parameters S_p, S_t for the parallel / transverse BAO smoothing.

The cost functions and Minuit drivers live in fitter (custom_least_squares*, cost_function, run_minuit, fitter_k_mu).

Conventions

• Wavenumber k in h/Mpc (change_k_normalization converts to/from 1/Mpc); mu = k_parallel / k.
• Power spectra are stored as P(k) (1D) or P(k, μ) (3D, with k_array stacked as [k, mu]).
• b and beta are the linear bias and redshift-space-distortion parameter.