For decades, the standard tool for understanding the universe has been the power spectrum. By measuring the "clumpiness" of matter at different scales, astronomers have been able to weigh the cosmos and chart its expansion. However, the power spectrum relies on a simplifying assumption: that the distribution of matter in the universe follows a Gaussian distribution.
While this was true in the very early universe, billions of years of gravity have changed the landscape. As Eiichiro Komatsu and colleagues argue in Non-Gaussianity as a Probe of the Physics of the Primordial Universe, the power spectrum acts as a lossy compression algorithm. It discards the "textures" of the universe—the filaments, voids, and complex shapes of the cosmic web—leaving significant cosmological information on the table.
The Information Leakage of Gravity
In the early universe, density fluctuations were small and random. But as the universe aged, gravity pulled matter into dense clusters and stretched it into long filaments, leaving behind vast, empty voids. This process is fundamentally non-linear.
As Francisco Villaescusa-Navarro and the team behind The Quijote simulations demonstrate, this non-linear evolution causes information to "leak" out of the two-point correlation functions (the basis of the power spectrum) and into higher-order patterns. If you only look at the power spectrum, you are essentially looking at a blurred map of the universe that tells you how much matter there is, but not how it is shaped.
To capture this missing information, researchers are moving toward higher-order statistics. For example, Zachary Slepian and Daniel J. Eisenstein achieved a milestone by detecting Baryon Acoustic Oscillations in the three-point correlation function. By looking at galaxy triplets rather than pairs, they could use the geometry of the cosmic web as a "standard ruler" with much higher precision.
Capturing Texture with Wavelets and Scattering Transforms
If the power spectrum measures the "volume" of cosmic structures, new mathematical tools are being developed to measure their "texture." One of the most promising is the Scattering Transform (ST).
As Sihao Cheng and Brice Ménard explain in A new approach to observational cosmology using the scattering transform, the ST functions like a "physicist’s neural network." It uses layers of wavelets to extract non-Gaussian features—like the sharp edges of filaments—without requiring the massive amounts of training data needed by standard deep learning.
Similarly, Erwan Allys and Michael Eickenberg have shown that 3D Wavelet Moments can provide parameter constraints 5 to 10 times tighter than the power spectrum. These tools work because they are sensitive to the phase of the density field—the specific alignment of matter that creates a filament rather than a random cloud.
Simulation-Based Inference and the Field-Level Approach
The challenge with non-linear regimes is that they are too complex for traditional "pen-and-paper" math. To solve this, researchers like ChangHoon Hahn and Shirley Ho are pioneering Simulation-Based Inference (SBI).
In the SimBIG framework, instead of trying to write an equation for the galaxy distribution, researchers "forward-model" the entire universe. They run thousands of N-body simulations with different physical constants and use machine learning to find which simulation best matches the raw, 3D "field" of observed galaxies.
By analyzing the raw field directly—rather than compressing it into a summary statistic—Pablo Lemos and the SimBIG team achieved constraints on the growth of structure ($\sigma_8$) that are more than double the precision of standard methods. This effectively "squeezes" more science out of existing telescope data, making a small survey perform like one twice its size.
Why It Works: Breaking Degeneracies
The ultimate goal of these non-Gaussian methods is to break parameter degeneracies. Often, two different versions of the universe—for example, one with more dark matter and one with heavier neutrinos—can produce identical power spectra.
However, as Elena Massara shows in her work on the Marked Power Spectrum, these two universes look very different if you focus on the voids. Neutrinos, being fast-moving "ghost particles," tend to smooth out the density in low-density regions. By using statistics that prioritize these "empty" spaces, researchers can finally "weigh" neutrinos and distinguish their effects from those of dark energy.
By moving beyond the Gaussian approximation, cosmology is transitioning from measuring the "average" universe to analyzing its specific, complex architecture. This shift promises to turn the "mildly non-linear" regime from a source of noise into the most information-rich frontier of modern physics.