Reading the early universe through the cosmic web

The universe’s earliest fluctuations left statistical clues in the later distribution of matter. Cosmologists search those clues for evidence about inflation, an early period of accelerated expansion. One target is primordial non-Gaussianity: departures from Gaussian statistics in the initial fluctuations.

A preprint by Kilian Peraza Vera, Francisco-Shu Kitaura and Francesco Sinigaglia, submitted on November 6, 2026, explores a different representation of one such signal. Instead of describing it only through clustering strength at different scales, the researchers describe it through the large-scale environments occupied by dark-matter halos. Their results come from simulations, not a detection in the observed universe.

Why the largest scales matter

The study concerns local-type primordial non-Gaussianity, whose amplitude is described by a parameter called f_NL. In this model, long-wavelength fluctuations influence shorter-wavelength fluctuations, changing the abundance and clustering of objects used to trace matter.

Those tracers do not follow matter’s distribution exactly. Cosmologists call the relationship between their clustering and matter’s clustering “bias.” Local-type non-Gaussianity produces a characteristic correction to that bias: on sufficiently large scales, it varies as f_NL times k to the power of minus two. Here k is a wavenumber; smaller k means a larger spatial scale.

That makes the largest scales particularly informative—and difficult. A finite survey contains relatively few enormous fluctuations, creating uncertainty known as cosmic variance. Survey systematics and effects associated with observing the universe can also contaminate the signal.

Four environments, four weights

The researchers construct a mathematical field from the smoothed matter density that strongly emphasizes long wavelengths. At each location, a tensor—a mathematical object encoding directional information—provides three eigenvalues. Counting how many exceed a threshold assigns that location to one of four classes: knot-like, filament-like, sheet-like or void-like.

These names require care. The classes describe broad, nonlocal environments, not simply the familiar smaller-scale knots and strands of the nonlinear cosmic web. “Nonlocal” means the classification depends on the surrounding density field rather than only the density at one point.

The construction’s key property is that the tensor’s trace, the sum of its diagonal components, carries the same inverse-square dependence on k as the leading primordial bias correction. The directional classification supplies a geometrical description of that response. It does not imply that the primordial universe had a preferred direction: the underlying model remains statistically isotropic.

The team assigns a weight to halos in each environment, then normalizes the weighted field to preserve its mean abundance. Four additional weights, derived from the standard tidal-field tensor, extend the model toward smaller spatial scales.

What the simulations showed

The authors tested the approach using AbacusPNG simulations at redshift 1.1, with f_NL values of −100, −30, 30 and 100. They fitted environmental weights using clustering measurements, including cross-power spectra that describe how halo and matter fluctuations vary together across scales.

The weighted Gaussian-reference catalogs reproduced the expected large-scale enhancement or suppression of bias in the non-Gaussian simulations. The authors report power-spectrum differences within the cosmic-variance uncertainty; the plotted uncertainty estimates use 20 Abacus simulations. Adding the tidal-environment weights improved agreement at higher wavenumbers, where the four-weight construction alone became insufficient.

The fitted weights followed a consistent ordering. For positive f_NL, knot-like environments received the greatest relative weight, followed by filaments, sheets and voids. For negative f_NL, the order reversed.

These are adjustments to a reference catalog, not evidence that ordinary Gaussian catalogs contain equal numbers of halos in every environment. Any observational comparison must subtract the appropriate Gaussian baseline.

A useful imitation, not a new cosmic history

The method could generate catalogs with an effective primordial clustering signature by reweighting existing halos or selectively retaining them. It does not move halos or recalculate their velocities. But changing their relative contributions or selecting a subset changes the resulting sample’s clustering statistics; random subsampling also introduces sampling noise.

Nor does the procedure recreate non-Gaussian merger histories or a fully evolved non-Gaussian halo mass function. The demonstrated test is a calibrated power-spectrum comparison. Independent realizations, intermediate f_NL values and primordial-signal-specific three-point statistics remain further tests.

For observations, the authors propose measuring environmental occupation contrasts alongside conventional clustering. They do not establish a precision improvement or eliminate uncertainty in how strongly a tracer population responds to primordial fluctuations. Survey boundaries, selection effects and redshift-space distortions still need dedicated treatment. Because the environmental reconstruction itself emphasizes long wavelengths, it is not immune to their contamination.

The result is therefore a simulation-tested representation of a known signal and a proposed complementary measurement—not evidence that primordial non-Gaussianity has been found.