Thursday.
1407.4118
Machine lingering classification of SDSS transient survey images
du Buisson et al
Show that multiple machine learning algorithms can match human performance in classifying transient imaging data from the SDSS SN survey into real objects and artifacts. This is the first step in any transient science pipeline and is currently still done by humans, but future surveys such as LSST will necessitate fully machine-enabled solutions. Using features trained from eigenimage analysis (PCA) of single-epoch g, r, i-difference images reaches a completeness (recall) of 95%, while only incorrectly classifying 18% of artifacts as real objects, corresponding to a precision (purity) of 85%. In general the k-nearest neighbor and the SkyNet ANN algorithms performed most robustly compared to other methods such as naive Bayes and kernel SVM. Results show that PCA-based machine learning can match human success levels and can naturally be extended by including multiple epochs of data, transient colors and host galaxy information which should allow for significant further improvements, especially at low signal to noise.
1407.4161
Simulation of a method to directly image exoplanets around multiple stars systems
Thomas, Belikov, Bendek
Direct imaging of extra-solar planets with specialized ground-based instruments (e.g. GPI) will allow detection of planets 1e7 times fainter than their host star. For space-based missions (e.g. WFIRST-AFTA), different teams have shown in laboratories contrasts reaching 1e-10 within a few diffraction limits from the star using a combination of a coronagraph to suppress light coming from the host star and wavefront control system. These demonstrations use a deformable mirror (DM) to remove residual starlight (speckles) created by the imperfections of telescope. However, all these current and future systems focus on detecting faint planets around a single host star or unresolved binaries/multiples, while several targets or planet candidates are located around nearby binary stars such as Alpha Centauri. Until now, it has been thought that removing the light of a companion star is impossible with the current technology, excluding binary star systems from target lists of direct imaging missions. Direct imaging around binaries or multiples systems at a level of contrast allowing Earth-like planets detection is challenging because the region of interest, where a dark zone is essential, is contaminated by the light coming from the host star's companion. Propose a method to simultaneously correct aberrations and diffraction of light coming from the target star. This method works even if the companion star is outside the control region of the DM (beyond its half-Nyquiest frequency), by taking advantage of aliasing effects.
1407.4233
Geometry of star-forming galaxies from SDSS, 3D-HST and CANDELS
van der Wel, ... van Dokkum, Franx, ... et al
Determine the intrinsic, 3d shape distribution of SF galaxies at 0<z<2.5, as inferred from their observed projected axis rations. In the present-day universe, SF galaxies of all masses 1e9-11 Msun are predominantly thin, nearly oblate disks, in line with previous studies. Now extend this to higher redshifts, and find that among massive galaxies (M*>1e10 Msun) disks are the most common geometric shape at all z<2. Lower-mass galaxies at z>1 possess a broad range of geometric shapes: the fraction of elongated (prolate) galaxies increases toward higher redshifts and lower masses. Galaxies with stellar mass 1e9 Msun (1e10 Msun) are a mix of roughly equal numbers of elongated and disk galaxies at z~1 (z~2). This suggests that galaxies in this mass range do not yet have disks that are sustained over many orbital periods, implying that galaxies with present-day stellar mass comparable to that of the MW typically first formed such sustained stellar disks at z~1.5-2. Combined with constraints on the evolution of the SFR density and the distribution of SF over galaxies with different masses, the findings imply that, averaged over cosmic time, the majority of stars formed in disks.
1407.4301
Baryons, Neutrinos, Feedback and Weak gravitational lensing
Harnois-Déraps, van Waerbeke, Viola, Heymans
The effect of baryonic feedback on the DM mass distribution is generally considered to be a nuisance to WL. Measurements of cosmological parameters are affected as feedback alters the cosmic shear signal on angular scales smaller than a few arc minutes. Recent progress on the numerical modeling of baryon physics has shown that this effect could be so large that, rather than being a nuisance, the effect can be constrained with current WL surveys, hence providing an alternative astrophysical insight on one of the most challenging questions of galaxy formation. In order to perform the analysis, construct an analytic fitting formula that describes the effect of the baryons on the mass PS. This fitting formula is based on 3 scenarios of the OWL hydrodynamical simulations. It is specifically calibrated for z<1.5, where it models the simulations to an accuracy that is better than 2% for scales k<10 h/Mpc and better tan 5% for 10<k<100 h/Mpc. Equipped with this precise tool, this paper present the first constraint on baryonic feedback models using gravitational lensing data, from the CFHTLenS. In this analysis, show that the effect of neutrino mass on the mass PS is degenerate with the baryonic feedback at small angular scales and cannot be ignored. Assuming a cosmology precision fixed by WMAP9, find that a universe with no baryon feedback and massless neutrinos is rejected by the CFHTLenS lensing data with 96% confidence. Study shows that ongoing WL surveys (KiDS, HSC and DES) will offer a unique opportunity to probe the physics of baryons at galactic scales, in addition to the expected constraints on the total neutrino mass.
1407.4354
The Galactic Center cloud G2 and its gas streamer
Pfuhl, et al
Detect preceding and lagging gas trails of G2. G1 preceded G2 by 13 years in the same orbit; G2 may be a bright knot in a much more extensive gas streamer. Imaging by SINFONI@VLT integral field spectroscopy.
No comments:
Post a Comment