Day 767

Monday.

1410.4559
Tow new methods to detect cosmic voids without density measurements
Elyiv, ... Branchini, etal

Two new void finders based on dynamical and clustering criteria to select voids in Lagrangian coordinates and minimize the impact of sparse sampling.  The first approach exploits the Zeldovich approximation to trace back in time the orbits of galaxies located in the voids and their surroundings, whereas the second uses the observed gg correlation function to relax the objects' spatial distribution to homogeneity and isotropy.  In both cases voids are defined as regions of the negative velocity divergence in Lagrangian coordinates, that can be regarded as sinks of the back-in-time streamlines of the mass tracers.  To assess the performance of the methods, use a DM halo catalogue extracted from an N-body simulation at z=0, and compared the results with those obtained with ZOBOV void finder.  Find that the void divergence profiles are less scattered than the density ones, so their stacking constitutes a more accurate cosmological probe.  The significance of the divergence signal in the central part of voids obtained from both finders is 60% higher than for overdensity profiles in the ZOBOV case.  Individual voids selected by both finders have similar tri-axial ellipsoidal shapes.  The ellipticity of the stacked void measured in the divergence field is significantly closer to unity, as expected, than what is found when using halo positions.  These results show that the new void finders are complementary to the existing methods, that should contribute to improve the accuracy of void-based cosmological tests.

1410.4565
Data mining for gravitationally lensed quasars
Agnello, Kelly, Treu, Marshall

A systematic exploration of machine learning techniques and demonstrate that a two step strategy can be highly effective.  In the first step, use catalog-level information (griz+WISE magnitudes, second moments) to preselect targets, using artificial neural networks.  The accepted targets are them inspected with pixel-by-pixel pattern recognition algorithms (Gradien-Boosted Trees), to form a final set of candidates.  The results from this procedure can be used to further refine the simpler SQLS algorithms, with a 2x (or 3x) gain in purity and the same (or 80%) completeness at target-selection stage, or a purity of 70% and a completeness of 60% after the candidate-selection step.  Simpler photometric searches in griz+WISE based on color cuts would provide samples with 7 purity or less.  The technique is extremely fast, as a list of candidates can be obtained from a stage III experiment (e.g. DES catalog/database) in a few CPU hours.  The techniques are easily extendable to Stage IV experiments like LSST with the addition of time domain information.

1410.4568
The mass-concentration relation in lensing clusters: the role of statistical biases and selection effects
Sereno, Giocoli, Ettori, Moscardini

The relation between mass and concentration of galaxy clusters traces their formation and evolution.  Massive lensing clusters were observed to be over-concentrated and following a steeper scaling in tension with predictions from standard LCDM.  Critically revise the relation in the CLASH, SGAS, LOCUSS and a high-z sample of WL clusters.  Measurements of mass and concentration are anti-correlated, which can bias the observed relation towards steeper values.  Corrected for this bias and compared the measured relation to theoretical predictions accounting for halo triaxiality, adiabatic contraction of the halo, presence of a dominant BCG and, mostly, select effects in the observed sample.  The normalization, the slope and the scatter of the expected relation are strongly sample-dependent.  For the considered samples, the predicted slope is much steeper than that of the underlying relation characterizing DM only cluster.  Find that correction of the statistical and selection biases mostly solve the tension with the LCDM model.

1410.4696
Feature importance for machine learning redshifts applied to SDSS galaxies
Hoyle, et al

Present an analysis of importance feature selection applied to photometric redshift estimation using the machine learning architecture Random Decision Forests (RDF) with the ensemble learning routing Adaboost.  Select a list of 85 easily measured (or derived) photometric quantities (or 'features') and spectroscopic redshifts for almost 2e6 galaxies from SDSS DR10.  After identifying which features have the most predictive power, use standard artificial Neural Networks (aNN) to show that the addition of these features, in combination with the standard magnitudes and colors, improves the machine learning redshift estimate by 18% and decrease the catastrophic outlier rate by 32%.  Further compare the redshift estimate from RDF using the ensemble learning routine Adaboost with those from two different aNNs, and with photometric redshifts available from the SDSS.  Find that the RDF requires orders of magnitude less computation time than the aNNs to obtain a machine learning redshift while reducing both the catastrophic outlier rate by put to 43%, and the redshift error by up to 25%.  When compared to the SDSS photo-z, the RDF machine learning redshifts both decreases the standard deviation of residuals scaled by 1/(1+z) by 36% from 0.066 to 0.041, and decreases the fraction of catastrophic outliers by 57% from 2.32% to 0.99%.