Neutrino mass constraint from an Implicit Likelihood Analysis of BOSS voids

Leander Thiele, Elena Massara, Alice Pisani, ChangHoon Hahn, David N. Spergel, Shirley Ho, B. D. Wandelt

TL;DR

This paper demonstrates how cosmic voids—the vast, underdense "empty" spaces between galaxy filaments—can be used to sharpen our measurements of neutrino masses. By applying simulation-based Implicit Likelihood Inference to BOSS survey data, the authors found that including void statistics tightens the upper bound on the sum of neutrino masses by approximately 20% compared to using galaxy clustering alone.

What Problem Does This Paper Address?

Neutrinos are among the most mysterious particles in the Standard Model. While we know they have mass due to oscillation experiments, we don't know the absolute scale (the sum of the masses of the three neutrino types, $\sum m_\nu$). Cosmology offers a unique laboratory for this: because neutrinos are light and move fast ("free-streaming"), they prevent matter from clumping together on small scales.

Most cosmological constraints come from the galaxy power spectrum (a two-point statistic measuring how galaxies cluster). However, this method misses "non-linear" information hidden in the distribution of matter. Cosmic voids are prime candidates for capturing this missing info because they are extremely underdense; while matter flows out of them into clusters, neutrinos are distributed more smoothly, making the neutrino-to-dark-matter ratio higher in voids. The challenge is that voids are difficult to model analytically, and traditional statistical methods (like Gaussian likelihoods) often fail to describe them accurately.

Methods

The researchers moved away from traditional analytical formulas and adopted a Simulation-Based Inference (SBI) approach, specifically Implicit Likelihood Inference (ILI).

  1. Simulations: They ran 127 cosmological simulations using FastPM (an approximate gravity solver) across a range of neutrino masses and cosmological parameters.
  2. Galaxy Mapping: They used an 11-parameter Halo Occupation Distribution (HOD) model to "paint" galaxies onto the dark matter halos, accounting for complex effects like assembly bias and velocity bias.
  3. Data Vector: They extracted three key statistics:
  4. Galaxy Auto Power Spectrum ($P^{gg}$): The standard clustering measure.
  5. Void-Galaxy Cross Power Spectrum ($P^{vg}$): How galaxies are distributed relative to void centers.
  6. Void Size Function (VSF): A count of voids of different sizes.
  7. Neural Ratio Estimation (NRE): Instead of assuming a likelihood shape, they trained a neural network to distinguish between "real" data-parameter pairs and "shuffled" ones. This allows the network to implicitly learn the likelihood-to-evidence ratio directly from the simulations.
  8. Compression: To make the high-dimensional data manageable for the neural network, they used MOPED (linear score compression) to reduce the data vector to just 17 summary statistics.

Datasets

  • BOSS CMASS NGC: The primary dataset is the "Constant Mass" (CMASS) sample from the North Galactic Cap (NGC) of the SDSS-III Baryon Oscillation Spectroscopic Survey.
  • Quijote Simulations: Used for preliminary fine-tuning of the HOD models.
  • Fiducial Mocks: 69 simulations with fixed cosmology used to validate the statistical independence of their data augmentations.

Key Insights and Ideas

  • Voids as Neutrino Probes: The study confirms that voids are not just "empty space" but are information-rich environments where the subtle effects of neutrino free-streaming are amplified.
  • Constraints "From Below": Interestingly, the authors found hints that void statistics might be particularly good at pushing the neutrino mass constraint up from zero, potentially helping to eventually detect a non-zero minimum mass rather than just lowering the ceiling.
  • Validating the Poisson Assumption: For years, cosmologists have assumed that void counts (the VSF) follow a Poisson distribution. This paper uses its large simulation suite to explicitly test and substantiate this assumption, providing a solid footing for future void studies.
  • Large Voids Carry the Signal: Most of the "neutrino signal" comes from relatively large voids (radii > 40 $h^{-1}$Mpc). This is a win for future surveys, as large voids are easier to identify even in noisier photometric data.

Key Results

  • Upper Bound: Using only the galaxy power spectrum ($k_{max}=0.15$ $h$ Mpc$^{-1}$), the 95% confidence limit on the neutrino mass sum was 0.43 eV.
  • The "Void Gain": When adding void statistics ($P^{vg}$ and VSF), the bound tightened to 0.35 eV—a ~20% improvement.
  • Robustness: The results held up under rigorous "coverage tests," proving that the neural network was well-trained and the resulting posteriors were statistically "calibrated" (i.e., they don't over- or under-estimate certainty).

Limitations and Future Work

  • Approximate Solvers: The use of FastPM is a trade-off for speed. While accurate on large scales, full N-body simulations would be required to push to smaller, more non-linear scales ($k > 0.2$).
  • Sample Size: The study only used the Northern part of the BOSS survey. Including the Southern Galactic Cap and the LOWZ (low redshift) samples would likely further tighten the bounds.
  • HOD Complexity: While the 11-parameter HOD is expressive, the authors note that more sophisticated modeling of how galaxies occupy halos (and how they are downsampled to match survey density) could reduce potential systematic biases.
  • Next-Gen Surveys: This framework is a "dress rehearsal" for upcoming surveys like DESI, Euclid, and the Vera Rubin Observatory (LSST), which will provide orders of magnitude more voids to analyze.

Summary updated

Abstract

Cosmic voids identified in the spatial distribution of galaxies provide complementary information to two-point statistics. In particular, constraints on the neutrino mass sum, $\sum m_ν$, promise to benefit from the inclusion of void statistics. We perform inference on the CMASS NGC sample of SDSS-III/BOSS with the aim of constraining $\sum m_ν$. We utilize the void size function, the void galaxy cross power spectrum, and the galaxy auto power spectrum. To extract constraints from these summary statistics we use a simulation-based approach, specifically implicit likelihood inference. We populate approximate gravity-only, particle neutrino cosmological simulations with an expressive halo occupation distribution model. With a conservative scale cut of kmax=0.15 h/Mpc and a Planck-inspired LCDM prior, we find upper bounds on $\sum m_ν$ of 0.43 and 0.35 eV from the galaxy auto power spectrum and the full data vector, respectively (95% credible interval). We observe hints that the void statistics may be most effective at constraining $\sum m_ν$ from below. We also substantiate the usual assumption that the void size function is Poisson distributed.
BibTeX
@article{thiele2024neutrino,
  author = {Leander Thiele and Elena Massara and Alice Pisani and ChangHoon Hahn and David N. Spergel and Shirley Ho and B. D. Wandelt},
  title = {{Neutrino mass constraint from an Implicit Likelihood Analysis of BOSS voids}},
  journal = {The Astrophysical Journal},
  year = {2024},
  doi = {10.3847/1538-4357/ad434e},
}
Figure