Weak-lensing mass reconstruction using sparsity and a Gaussian random field

Aims. We introduce a novel approach to reconstructing dark matter mass maps from weak gravitational lensing measurements. The cornerstone of the proposed method lies in a new modelling of the matter density field in the Universe as a mixture of two components: (1) a sparsity-based component that captures the non-Gaussian structure of the field, such as peaks or halos at different spatial scales, and (2) a Gaussian random field, which is known to represent the linear characteristics of the field well.Methods. We propose an algorithm called MCALens that jointly estimates these two components. MCALens is based on an alternating minimisation incorporating both sparse recovery and a proximal iterative Wiener filtering.Results. Experimental results on simulated data show that the proposed method exhibits improved estimation accuracy compared to customised mass-map reconstruction methods.

Methods for strong gravitational lens detection and analysis using machine learning and high performance computing

Christoph Ernst René Schäfer

The study of strong gravitational lenses is a relatively new scientific field in astronomy with many applications in cosmology. Its unique observables allow astronomers to trace dark matter, determine the expansion of the universe and study galaxy evolution. While strong lenses are relatively rare phenomena, recent and future improvements in observational instrumentation are expected to lead to a sharp increase in the number and quality of their observation. With that increase comes however an observational challenge from the sheer quantity of data the lenses are hidden in. Current classification and modelling techniques are simply not capable of handling the new data in a reasonable timeframe. As a consequence this thesis concerns itself with the development and improvement of numerical methods necessary to find and analyse these newly found lenses, borrowing from deep learning and high performance computing techniques. Deep learning, a field of machine learning, shows significant success in finding lenses in simulations. Performing significantly better than any other classification method, CNN-based lens finders reach almost the required classification efficiency and accuracy for a fully automated Euclid strong lensing pipeline. Its dependency on a varied and complete training set and the difficulty creating it make its immediate application to new surveys however more difficult. Incorporating high performance computing techniques and more concurrent algorithm design into lens analysis results in crucial speed-ups for strong lens analysis. Applied to Lenstool, a popular mass modelling tool for gravitational lenses, the resulting software can be run on modern supercomputers, using Graphics Processing Units (GPU) and CPUs simultaneously. Running it on state-of-the-art hardware makes it possible to reduce computation time to an acceptable level when mass modelling complex cluster lenses.

EPFL2020

Modelling Small-Scale Drifting Snow with a Lagrangian Stochastic Model Based on Large-Eddy Simulations

Marc Diebold, Michael Lehning, Gian-Duri Lieberherr, Jan Overney, Marc Parlange

Observations of drifting snow on small scales have shown that, in spite of nearly steady winds, the snow mass flux can strongly fluctuate in time and space. Most drifting snow models, however, are not able to describe drifting snow accurately over short time periods or on small spatial scales as they rely on mean flow fields and assume equilibrium saltation. In an attempt to gain understanding of the temporal and spatial variability of drifting snow on small scales, we propose to use a model combination of flow fields from large-eddy simulations (LES) and a Lagrangian stochastic model to calculate snow particle trajectories and so infer snow mass fluxes. Model results show that, if particle aerodynamic entrainment is driven by the shear stress retrieved from the LES, we can obtain a snow mass flux varying in space and time. The obtained fluctuating snow mass flux is qualitatively compared to field and wind-tunnel measurements. The comparison shows that the model results capture the intermittent behaviour of observed drifting snow mass flux yet differences between modelled turbulent structures and those likely to be found in the field complicate quantitative comparisons. Results of a model experiment show that the surface shear-stress distribution and its influence on aerodynamic entrainment appear to be key factors in explaining the intermittency of drifting snow.

Springer Verlag2014

Hopfield neural network deconvolution for weak lensing measurement

Frédéric Courbin, Georges Meylan, Guldariya Nurbaeva, Malte Tewes

Weak gravitational lensing has the potential to place tight constraints on the equation of the state of dark energy. However, this will only be possible if shear measurement methods can reach the required level of accuracy. We present a new method for measuring the ellipticity of galaxies used in weak lensing surveys. The method makes use of direct deconvolution of the data by the total point spread function (PSF). We adopt a linear algebra formalism that represents the PSF as a Toeplitz matrix. This allows us to solve the convolution equation by applying the Hopfield neural network iterative scheme. The ellipticity of galaxies in the deconvolved images are then measured using second-order moments of the autocorrelation function of the images. To our knowledge, it is the first time full image deconvolution has been used to measure weak lensing shear. We apply our method to the simulated weak lensing data proposed in the GREAT10 challenge and obtain a quality factor of Q = 87. This result is obtained after applying image denoising to the data, prior to the deconvolution. The additive and multiplicative biases on the shear power spectrum are then root A = +0.09 x 10(-4) and M/2 = +0.0357, respectively.

EDP Sciences2015

Methods for strong gravitational lens detection and analysis using machine learning and high performance computing

Christoph Ernst René Schäfer

EPFL2020