Search for: All records

In this paper, we present work towards the development of a new data analytics and machine learning (ML) framework, called MagmaDNN. Our main goal is to provide scalable, high-performance data analytics and ML solutions for scientific applications running on current and upcoming heterogeneous many-core GPU-accelerated architectures. To this end, since many of the functionalities needed are based on standard linear algebra (LA) routines, we designed MagmaDNN to derive its performance power from the MAGMA library. The close integration provides the fundamental (scalable high-performance) LA routines available in MAGMA as a backend to MagmaDNN. We present some design issues for performance and scalability that are specific to ML using Deep Neural Networks (DNN), as well as the MagmaDNN designs towards overcoming them. In particular, MagmaDNN uses well established HPC techniques from the area of dense LA, including task-based parallelization, DAG representations, scheduling, mixed-precision algorithms, asynchronous solvers, and autotuned hyperparameter optimization. We illustrate these techniques and their incorporation and use to outperform other frameworks, currently available.

Abstract We investigate the environment and line of sight of the H0LiCOW lens B1608+656 using Subaru Suprime-Cam and the Hubble Space Telescope (HST) to perform a weak lensing analysis. We compare three different methods to reconstruct the mass map of the field, i.e. the standard Kaiser-Squires inversion coupled with inpainting and Gaussian or wavelet filtering, and ${\tt Glimpse}$ a method based on sparse regularization of the shear field. We find no substantial difference between the 2D mass reconstructions, but we find that the ground-based data is less sensitive to small-scale structures than the space-based observations. Marginalising over the results obtained with all the reconstruction techniques applied to the two available HST filters F606W and F814W, we estimate the external convergence, κext at the position of B1608+656 is $\kappa _{\mathrm{ext}}= 0.11^{+0.06}_{-0.04}$, where the error bars corresponds respectively to the 16th and 84th quartiles. This result is compatible with previous estimates using the number-counts technique, suggesting that B1608+656 resides in an over-dense line of sight, but with a completely different technique. Using our mass reconstructions, we also compare the convergence at the position of several groups of galaxies in the field of B1608+656 with the mass measurements using various analytical mass profiles,more »and find that the weak lensing results favor truncated halo models.« less

The H0LiCOW collaboration inferred via strong gravitational lensing time delays a Hubble constant value of H 0 = 73.3 −1.8 +1.7 km s −1 Mpc −1 , describing deflector mass density profiles by either a power-law or stars (constant mass-to-light ratio) plus standard dark matter halos. The mass-sheet transform (MST) that leaves the lensing observables unchanged is considered the dominant source of residual uncertainty in H 0 . We quantify any potential effect of the MST with a flexible family of mass models, which directly encodes it, and they are hence maximally degenerate with H 0 . Our calculation is based on a new hierarchical Bayesian approach in which the MST is only constrained by stellar kinematics. The approach is validated on mock lenses, which are generated from hydrodynamic simulations. We first applied the inference to the TDCOSMO sample of seven lenses, six of which are from H0LiCOW, and measured H 0 = 74.5 −6.1 +5.6 km s −1 Mpc −1 . Secondly, in order to further constrain the deflector mass density profiles, we added imaging and spectroscopy for a set of 33 strong gravitational lenses from the Sloan Lens ACS (SLACS) sample. For nine of the 33 SLAC lenses,more »we used resolved kinematics to constrain the stellar anisotropy. From the joint hierarchical analysis of the TDCOSMO+SLACS sample, we measured H 0 = 67.4 −3.2 +4.1 km s −1 Mpc −1 . This measurement assumes that the TDCOSMO and SLACS galaxies are drawn from the same parent population. The blind H0LiCOW, TDCOSMO-only and TDCOSMO+SLACS analyses are in mutual statistical agreement. The TDCOSMO+SLACS analysis prefers marginally shallower mass profiles than H0LiCOW or TDCOSMO-only. Without relying on the form of the mass density profile used by H0LiCOW, we achieve a ∼5% measurement of H 0 . While our new hierarchical analysis does not statistically invalidate the mass profile assumptions by H0LiCOW – and thus the H 0 measurement relying on them – it demonstrates the importance of understanding the mass density profile of elliptical galaxies. The uncertainties on H 0 derived in this paper can be reduced by physical or observational priors on the form of the mass profile, or by additional data.« less

Time-delay cosmography of lensed quasars has achieved 2.4% precision on the measurement of the Hubble constant, H 0 . As part of an ongoing effort to uncover and control systematic uncertainties, we investigate three potential sources: 1- stellar kinematics, 2- line-of-sight effects, and 3- the deflector mass model. To meet this goal in a quantitative way, we reproduced the H0LiCOW/SHARP/STRIDES (hereafter TDCOSMO) procedures on a set of real and simulated data, and we find the following. First, stellar kinematics cannot be a dominant source of error or bias since we find that a systematic change of 10% of measured velocity dispersion leads to only a 0.7% shift on H 0 from the seven lenses analyzed by TDCOSMO. Second, we find no bias to arise from incorrect estimation of the line-of-sight effects. Third, we show that elliptical composite (stars + dark matter halo), power-law, and cored power-law mass profiles have the flexibility to yield a broad range in H 0 values. However, the TDCOSMO procedures that model the data with both composite and power-law mass profiles are informative. If the models agree, as we observe in real systems owing to the “bulge-halo” conspiracy, H 0 is recovered precisely and accurately bymore »both models. If the two models disagree, as in the case of some pathological models illustrated here, the TDCOSMO procedure either discriminates between them through the goodness of fit, or it accounts for the discrepancy in the final error bars provided by the analysis. This conclusion is consistent with a reanalysis of six of the TDCOSMO (real) lenses: the composite model yields H 0 = 74.0 −1.8 +1.7 km s −1 Mpc −1 , while the power-law model yields 74.2 −1.6 +1.6 km s −1 Mpc −1 . In conclusion, we find no evidence of bias or errors larger than the current statistical uncertainties reported by TDCOSMO.« less

We present new measurements of the time delays of WFI2033−4723. The data sets used in this work include 14 years of data taken at the 1.2 m Leonhard Euler Swiss telescope, 13 years of data from the SMARTS 1.3 m telescope at Las Campanas Observatory and a single year of high-cadence and high-precision monitoring at the MPIA 2.2 m telescope. The time delays measured from these different data sets, all taken in the R -band, are in good agreement with each other and with previous measurements from the literature. Combining all the time-delay estimates from our data sets results in Δ t AB = 36.2 +0.7 −0.8 days (2.1% precision), Δ t AC = −23.3 +1.2 −1.4 days (5.6%) and Δ t BC = −59.4 +1.3 −1.3 days (2.2%). In addition, the close image pair A1-A2 of the lensed quasars can be resolved in the MPIA 2.2 m data. We measure a time delay consistent with zero in this pair of images. We also explore the prior distributions of microlensing time-delay potentially affecting the cosmological time-delay measurements of WFI2033−4723. Our time-delay measurements are not precise enough to conclude that microlensing time delay is present or absent from the data. Thismore »work is part of a H0LiCOW series focusing on measuring the Hubble constant from WFI2033−4723.« less

ABSTRACT In time-delay cosmography, three of the key ingredients are (1) determining the velocity dispersion of the lensing galaxy, (2) identifying galaxies and groups along the line of sight with sufficient proximity and mass to be included in the mass model, and (3) estimating the external convergence κext from less massive structures that are not included in the mass model. We present results on all three of these ingredients for two time-delay lensed quad quasar systems, DES J0408–5354 and WGD 2038–4008 . We use the Gemini, Magellan, and VLT telescopes to obtain spectra to both measure the stellar velocity dispersions of the main lensing galaxies and to identify the line-of-sight galaxies in these systems. Next, we identify 10 groups in DES J0408–5354 and two groups in WGD 2038–4008 using a group-finding algorithm. We then identify the most significant galaxy and galaxy-group perturbers using the ‘flexion shift’ criterion. We determine the probability distribution function of the external convergence κext for both of these systems based on our spectroscopy and on the DES-only multiband wide-field observations. Using weighted galaxy counts, calibrated based on the Millennium Simulation, we find that DES J0408–5354 is located in a significantly underdense environment, leading to a tight (width $\sim 3{{\ \rm per\ cent}}$), negative-value κext distribution. Onmore »the other hand, WGD 2038–4008 is located in an environment of close to unit density, and its low source redshift results in a much tighter κext of $\sim 1{{\ \rm per\ cent}}$, as long as no external shear constraints are imposed.« less

ABSTRACT Galaxies and galaxy groups located along the line of sight towards gravitationally lensed quasars produce high-order perturbations of the gravitational potential at the lens position. When these perturbation are too large, they can induce a systematic error on H0 of a few per cent if the lens system is used for cosmological inference and the perturbers are not explicitly accounted for in the lens model. In this work, we present a detailed characterization of the environment of the lens system WFI 2033−4723 ($z_{\rm src} =\,$1.662, $z_{\rm lens}=\,$0.6575), one of the core targets of the H0LiCOW project for which we present cosmological inferences in a companion paper. We use the Gemini and ESO-Very Large telescopes to measure the spectroscopic redshifts of the brightest galaxies towards the lens, and use the ESO-MUSE integral field spectrograph to measure the velocity-dispersion of the lens ($\sigma _{\rm {los}}= 250^{+15}_{-21}$  km s−1) and of several nearby galaxies. In addition, we measure photometric redshifts and stellar masses of all galaxies down to i < 23 mag, mainly based on Dark Energy Survey imaging (DR1). Our new catalogue, complemented with literature data, more than doubles the number of known galaxy spectroscopic redshifts in the direct vicinity of the lens, expanding to 116more »(64) the number of spectroscopic redshifts for galaxies separated by less than 3 arcmin (2 arcmin ) from the lens. Using the flexion-shift as a measure of the amplitude of the gravitational perturbation, we identify two galaxy groups and three galaxies that require specific attention in the lens models. The ESO MUSE data enable us to measure the velocity-dispersions of three of these galaxies. These results are essential for the cosmological inference analysis presented in Rusu et al.« less