Gaussianisation for fast and accurate inference from cosmological data

Abstract

We present a method to transform multivariate unimodal non-Gaussian posterior probability densities into approximately Gaussian ones via non-linear mappings, such as Box--Cox transformations and generalisations thereof. This permits an analytical reconstruction of the posterior from a point sample, like a Markov chain, and simplifies the subsequent joint analysis with other experiments. This way, a multivariate posterior density can be reported efficiently, by compressing the information contained in MCMC samples. Further, the model evidence integral (i.e. the marginal likelihood) can be computed analytically. This method is analogous to the search for normal parameters in the cosmic microwave background, but is more general. The search for the optimally Gaussianising transformation is performed computationally through a maximum-likelihood formalism; its quality can be judged by how well the credible regions of the posterior are reproduced. We demonstrate that our method outperforms kernel density estimates in this objective. Further, we select marginal posterior samples from Planck data with several distinct strongly non-Gaussian features, and verify the reproduction of the marginal contours. To demonstrate evidence computation, we Gaussianise the joint distribution of data from weak lensing and baryon acoustic oscillations (BAO), for different cosmological models, and find a preference for flat $\Lambda$CDM. Comparing to values computed with the Savage-Dickey density ratio, and Population Monte Carlo, we find good agreement of our method within the spread of the other two.

14 pages, 9 figures

References in corpus (3)

• The clustering of galaxies in the SDSS-III Baryon Oscillation Spectroscopic Survey: Baryon Acoustic Oscillations in the Data Release 10 and 11 galaxy samples
• Measuring the Baryon Acoustic Oscillation scale using the SDSS and 2dFGRS
• Pico: Parameters for the Impatient Cosmologist

Cited by in corpus (20)

• Baryon Acoustic Oscillations and the Hubble Constant: Past, Present and Future
• Improving lognormal models for cosmological fields
• Quantifying tensions in cosmological parameters: Interpreting the DES evidence ratio
• Assessing tension metrics with Dark Energy Survey and Planck data
• Maximal compression of the redshift space galaxy power spectrum and bispectrum
• Enhancing BOSS bispectrum cosmological constraints with maximal compression
• Quantifying Tensions between CMB and Distance Datasets in Models with Free Curvature or Lensing Amplitude
• Quantifying Suspiciousness Within Correlated Data Sets
• A new model to predict weak-lensing peak counts III. Filtering technique comparisons
• Cosmological Inference from Galaxy-Clustering Power Spectrum: Gaussianization and Covariance Decomposition
• When tension is just a fluctuation: How noisy data affect model comparison
• Integrated cosmological probes: Concordance quantified
• Posterior inference unchained with EL_2O
• Information geometry in cosmological inference problems
• Higher Order Hamiltonian Monte Carlo Sampling for Cosmological Large-Scale Structure Analysis
• Deep Learning and genetic algorithms for cosmological Bayesian inference speed-up
• Data-Space Validation of High-Dimensional Models by Comparing Sample Quantiles
• Accounting for Correlations When Fitting Extra Cosmological Parameters
• Cross-correlations between mm-wave line-intensity mapping and weak lensing surveys: preliminary consideration of long-term prospects
• The Kaiser-Rocket effect: three decades and counting