Skip to content

SIGWAY

SIGWAY SIGWAY

SIGWAY (Scalar-Induced Gravitational Wave AnalYsis) computes the stochastic gravitational-wave background induced at second order by primordial curvature perturbations — scalar-induced gravitational waves (SIGWs). It is under active development by the LISA Cosmology Working Group; if you use it, please cite arXiv:2501.11320.

You build a spectrum from three interchangeable pieces, composed into one model:

flowchart LR
  P["𝒫<sub>ζ</sub>(k)<br/>primordial spectrum"] --> G(["OmegaGW"])
  K["kernel<br/>source era"] --> G
  I["integrator<br/>(s, t) quadrature"] --> G
  G --> O["Ω<sub>GW</sub>(f)"]

The whole pipeline is written in JAX: every model is jit-compiled and differentiable, so a Fisher forecast is one .jacobian call away.

What it can do

  • Analytic power spectra. Hand AnalyticPerturbations any closed-form \(\mathcal{P}_\zeta(k)\) — a log-normal peak, broken power law, flat-with-cutoff, oscillatory multifield template, … — and get \(\Omega_{\mathrm{GW}}(f)\). Cheap and fully differentiable.
  • Inflation from first principles. SingleFieldPerturbations wraps the Mukhanov–Sasaki solver (SingleFieldSolver): give it an inflaton potential \(V(\phi)\) and it integrates the background and the mode equations to produce \(\mathcal{P}_\zeta\) — e.g. ultra-slow-roll quasi-inflection-point models.
  • Binned / model-independent reconstruction. Binned_P_zeta represents \(\mathcal{P}_\zeta\) as free amplitudes in \(k\)-bins — a template-free way to fit or forecast the spectrum, exposing the same parameter_names / jacobian interface as everything else.
  • Different cosmological eras. Switch the source era by swapping the kernel: RadiationKernel (radiation domination) or InstantEMDKernel (an early matter-dominated era ending in a sudden transition to radiation).
  • Built for inference. Models are jit-compiled and differentiable: one ordered parameter vector plus a .jacobian (forward-mode autodiff, with finite differences for non-smooth knobs such as a source cutoff) gives Fisher forecasts directly.
  • Composable & extensible. OmegaGW simply composes a perturbations object, a kernel and an integrator — each a small base class you can subclass to add new physics or a new quadrature.

Where to start