Search

We build a deep learning framework that connects the local formation process of dark matter halos to the halo bias. We train a convolutional neural network (CNN) to predict the final mass and concentration of dark matter halos from the initial conditions. The CNN is then used as a surrogate model to derive the response of the halos' mass and concentration to long-wavelength perturbations in the initial conditions, and consequently the halo bias parameters following the "response bias" definition. The CNN correctly predicts how the local properties of dark matter halos respond to changes in the large-scale environment, despite no explicit knowledge of halo bias being provided during training. We show that the CNN recovers the known trends for the linear and second-order density bias parameters $b_1$ and $b_2$, as well as for the local primordial non-Gaussianity linear bias parameter $b_\phi$. The expected secondary assembly bias dependence on halo concentration is also recovered by the CNN: at fixed mass, halo concentration has only a mild impact on $b_1$, but a strong impact on $b_\phi$. Our framework opens a new window for discovering which physical aspects of the halo's Lagrangian patch determine assembly bias, which in turn can inform physical models of halo formation and bias., Comment: 11 pages, 5 figures, to be submitted to MNRAS, comments welcome

We discuss the potential of the multi-tracer technique to improve observational constraints of the local primordial non-Gaussianity (PNG) parameter $f_{\rm NL}$ from the galaxy power spectrum. For two galaxy samples $A$ and $B$, we show the constraining power is $\propto |b_1^B b_\phi^A - b_1^A b_\phi^B|$, where $b_1$ and $b_\phi$ are the linear and PNG galaxy bias parameters. This allows for significantly improved constraints compared to the traditional expectation $\propto |b_1^A - b_1^B|$ based on naive universality-like relations where $b_\phi \propto b_1$. Using IllustrisTNG galaxy simulation data, we find that different equal galaxy number splits of the full sample lead to different $|b_1^B b_\phi^A - b_1^A b_\phi^B|$, and thus have different constraining power. Of all of the strategies explored, splitting by $g-r$ color is the most promising, more than doubling the significance of detecting $f_{\rm NL}b_\phi \neq 0$. Importantly, since these are constraints on $f_{\rm NL}b_\phi$ and not $f_{\rm NL}$, they do not require priors on the $b_\phi(b_1)$ relation. For direct constraints on $f_{\rm NL}$, we show that multi-tracer constraints can be significantly more robust than single-tracer to $b_\phi$ misspecifications and uncertainties; this relaxes the precision and accuracy requirements for $b_\phi$ priors. Our results present new opportunities to improve our chances to detect and robustly constrain $f_{\rm NL}$, and strongly motivate galaxy formation simulation campaigns to calibrate the $b_\phi(b_1)$ relation., Comment: 9 pages, 4 figures, 1 table. Comments welcomed!

We study the effect of super-sample covariance (SSC) on the power spectrum and higher-order statistics: bispectrum, halo mass function, and void size function. We also investigate the effect of SSC on the cross-covariance between the statistics. We consider both the matter and halo fields. Higher-order statistics of the large-scale structure contain additional cosmological information beyond the power spectrum and are a powerful tool to constrain cosmology. They are a promising probe for ongoing and upcoming high precision cosmological surveys such as DESI, PFS, Rubin Observatory LSST, Euclid, SPHEREx, SKA, and Roman Space Telescope. Cosmological simulations used in modeling and validating these statistics often have sizes that are much smaller than the observed Universe. Density fluctuations on scales larger than the simulation box, known as super-sample modes, are not captured by the simulations and in turn can lead to inaccuracies in the covariance matrix. We compare the covariance measured using simulation boxes containing super-sample modes to those without. We also compare with the Separate Universe approach. We find that while the power spectrum, bispectrum and halo mass function show significant scale- or mass-dependent SSC, the void size function shows relatively small SSC. We also find significant SSC contributions to the cross-covariances between the different statistics, implying that future joint-analyses will need to carefully take into consideration the effect of SSC., 8+2 pages, 4 figures

We show that, in the presence of massive neutrinos, the Galileon gravity model provides a very good fit to the current CMB temperature, CMB lensing and BAO data. This model, which we dub ${\nu} \rm{Galileon}$, when assuming its stable attractor background solution, contains the same set of free parameters as $\Lambda\rm{CDM}$, although it leads to different expansion dynamics and nontrivial gravitational interactions. The data provide compelling evidence ($\gtrsim 6\sigma$) for nonzero neutrino masses, with $\Sigma m_\nu \gtrsim 0.4\ {\rm eV}$ at the $2\sigma$ level. Upcoming precision terrestrial measurements of the absolute neutrino mass scale therefore have the potential to test this model. We show that CMB lensing measurements at multipoles $l \lesssim 40$ will be able to discriminate between the ${\nu} \rm{Galileon}$ and $\Lambda\rm{CDM}$ models. Unlike $\Lambda\rm{CDM}$, the ${\nu} \rm{Galileon}$ model is consistent with local determinations of the Hubble parameter. The presence of massive neutrinos lowers the value of $\sigma_8$ substantially, despite of the enhanced gravitational strength on large scales. Unlike $\Lambda\rm{CDM}$, the ${\nu} \rm{Galileon}$ model predicts a negative ISW effect, which is difficult to reconcile with current observational limits., Comment: 4 pages, 2 figures, 1 table. Version 3: matches published version