Work

Stabilizing Fisher Forecasts for Power Spectrum and Bispectrum

Year: 2026
Type: preprint
Abstract: Fisher forecasts are a standard tool for assessing how well a summary statistic can constrain cosmological parameters, but they become unreliable when the derivatives of the statistic are estimated from a limited number of simulations. We apply a recently proposed stability framework, which selects informative and stable bins of a summary statistic, to the matter power spectrum and the bispectrum monopole measured in the Quijote simulations, and perform individual and combined Fisher forecasts for Ωm and ns. Restricted to 13 bins per statistic, the bispectrum constrains both parameters more weakly than the power spectrum, although its forecasts converge faster with the number of realisations. Combining the two under a block-diagonal covariance, which neglects their cross-covariance and is therefore likely optimistic, tightens the power spectrum constraints by about 40% on Ωm and 25% on ns, and this result is stable to the choice of selection cuts. Measured against a held-out reference set, the selected bins reach a given forecast stability with substantially fewer realisations than randomly chosen bins (≳ 20 times fewer for the power spectrum, about 2 times for the bispectrum), and forecasts from 100 derivative realisations agree with those from 400 further realisations to within about 5%.
Language: English
Open Access status: green
OpenAlex ID: W7225897218
Cites:
Cited by:
Related to: 10