Search for: All records

We introduce a novel approach to boost the efficiency of the importance nested sampling (INS) technique for Bayesian posterior and evidence estimation using deep learning. Unlike rejection-based sampling methods such as vanilla nested sampling (NS) or Markov chain Monte Carlo (MCMC) algorithms, importance sampling techniques can use all likelihood evaluations for posterior and evidence estimation. However, for efficient importance sampling, one needs proposal distributions that closely mimic the posterior distributions. We show how to combine INS with deep learning via neural network regression to accomplish this task. We also introduce nautilus, a reference open-source python implementation of this technique for Bayesian posterior and evidence estimation. We compare nautilus against popular NS and MCMC packages, including emcee, dynesty, ultranest, and pocomc, on a variety of challenging synthetic problems and real-world applications in exoplanet detection, galaxy SED fitting and cosmology. In all applications, the sampling efficiency of nautilus is substantially higher than that of all other samplers, often by more than an order of magnitude. Simultaneously, nautilus delivers highly accurate results and needs fewer likelihood evaluations than all other samplers tested. We also show that nautilus has good scaling with the dimensionality of the likelihood and is easily parallelizable to many CPUs.

 

We present a novel simulation-based cosmological analysis of galaxy–galaxy lensing and galaxy redshift-space clustering. Compared to analysis methods based on perturbation theory, our simulation-based approach allows us to probe a much wider range of scales, $0.4 \, h^{-1} \, \mathrm{Mpc}$ to $63 \, h^{-1} \, \mathrm{Mpc}$, including highly non-linear scales, and marginalizes over astrophysical effects such as assembly bias. We apply this framework to data from the Baryon Oscillation Spectroscopic Survey LOWZ sample cross-correlated with state-of-the-art gravitational lensing catalogues from the Kilo Degree Survey and the Dark Energy Survey. We show that gravitational lensing and redshift-space clustering when analysed over a large range of scales place tight constraints on the growth-of-structure parameter $S_8 = \sigma _8 \sqrt{\Omega _{\rm m} / 0.3}$. Overall, we infer S8 = 0.792 ± 0.022 when analysing the combination of galaxy–galaxy lensing and projected galaxy clustering and S8 = 0.771 ± 0.027 for galaxy redshift-space clustering. These findings highlight the potential constraining power of full-scale studies over studies analysing only large scales and also showcase the benefits of analysing multiple large-scale structure surveys jointly. Our inferred values for S8 fall below the value inferred from the CMB, S8 = 0.834 ± 0.016. While this difference is not statistically significant by itself, our results mirror other findings in the literature whereby low-redshift large-scale structure probes infer lower values for S8 than the CMB, the so-called S8-tension.

 

We present observational constraints on the galaxy–halo connection, focusing particularly on galaxy assembly bias from a novel combination of counts-in-cylinders statistics, P(NCIC), with the standard measurements of the projected two-point correlation function wp(rp), and number density ngal of galaxies. We measure ngal, wp(rp), and P(NCIC) for volume-limited, luminosity-threshold samples of galaxies selected from SDSS DR7, and use them to constrain halo occupation distribution (HOD) models, including a model in which galaxy occupation depends upon a secondary halo property, namely halo concentration. We detect significant positive central assembly bias for the Mr < −20.0 and Mr < −19.5 samples. Central galaxies preferentially reside within haloes of high concentration at fixed mass. Positive central assembly bias is also favoured in the Mr < −20.5 and Mr < −19.0 samples. We find no evidence of central assembly bias in the Mr < −21.0 sample. We observe only a marginal preference for negative satellite assembly bias in the Mr < −20.0 and Mr < −19.0 samples, and non-zero satellite assembly bias is not indicated in other samples. Our findings underscore the necessity of accounting for galaxy assembly bias when interpreting galaxy survey data, and demonstrate the potential of count statistics in extracting information from the spatial distribution of galaxies, which could be applied to both galaxy–halo connection studies and cosmological analyses.

 

ABSTRACT We use a simulation-based modelling approach to analyse the anisotropic clustering of the BOSS LOWZ sample over the radial range $0.4 \, h^{-1} \, \mathrm{Mpc}$ to $63 \, h^{-1} \, \mathrm{Mpc}$, significantly extending what is possible with a purely analytic modelling framework. Our full-scale analysis yields constraints on the growth of structure that are a factor of two more stringent than any other study on large scales at similar redshifts. We infer fσ8 = 0.471 ± 0.024 at $z$ ≈ 0.25, and fσ8 = 0.430 ± 0.025 at $z$ ≈ 0.40; the corresponding ΛCDM predictions of the Planck cosmic microwave background (CMB) analysis are 0.470 ± 0.006 and 0.476 ± 0.005, respectively. Our results are thus consistent with Planck, but also follow the trend seen in previous low-redshift measurements of fσ8 falling slightly below the ΛCDM + CMB prediction. We find that small- and large-radial scales yield mutually consistent values of fσ8, but there are 1−2.5σ hints of small scales ($\lt 10 \, h^{-1} \, \mathrm{Mpc}$) preferring lower values for fσ8 relative to larger scales. We analyse the constraining power of the full range of radial scales, finding that most of the multipole information about fσ8 is contained in the scales $2 \, h^{-1} \, \mathrm{Mpc}\lesssim s \lesssim 20 \, h^{-1} \, \mathrm{Mpc}$. Evidently, once the cosmological information of the quasi-to-nonlinear regime has been harvested, large-scale modes contain only modest additional information about structure growth. Finally, we compare predictions for the galaxy–galaxy lensing amplitude of the two samples against measurements from SDSS and assess the lensing-is-low effect in light of our findings. 

ABSTRACT The canonical Lambda cold dark matter (ΛCDM) cosmological model makes precise predictions for the clustering and lensing properties of galaxies. It has been shown that the lensing amplitude of galaxies in the Baryon Oscillation Spectroscopic Survey (BOSS) is lower than expected given their clustering properties. We present new measurements and modelling of galaxies in the BOSS LOWZ sample. We focus on the radial and stellar mass dependence of the lensing amplitude mismatch. We find an amplitude mismatch of around $35{{\ \rm per\ cent}}$ when assuming ΛCDM with Planck Cosmological Microwave Background (CMB) constraints. This offset is independent of halo mass and radial scale in the range Mhalo ∼ 1013.3−1013.9h−1 M⊙ and $r=0.1\!-\!60 \, h^{-1} \mathrm{Mpc}$ ($k \approx 0.05\!-\!20 \, h \, {\rm Mpc}^{-1}$). The observation that the offset is both mass and scale independent places important constraints on the degree to which astrophysical processes (baryonic effects, assembly bias) can fully explain the effect. This scale independence also suggests that the ‘lensing is low’ effect on small and large radial scales probably have the same physical origin. Resolutions based on new physics require a nearly uniform suppression, relative to ΛCDM predictions, of the amplitude of matter fluctuations on these scales. The possible causes of this are tightly constrained by measurements of the CMB and of the low-redshift expansion history.