24 May 2019 in Monthly Notices of the Royal Astronomical Society
JCJean-Charles Cuillandre
+1
18 October 2019 in Monthly Notices of the Royal Astronomical Society
Machine learning is becoming a popular tool to quantify galaxy morphologies and identify mergers. However, this technique relies on using an appropriate set of training data to be successful. By combining hydrodynamical simulations, synthetic observations, and convolutional neural networks (CNNs), we quantitatively assess how realistic simulated galaxy images must be in order to reliably classify mergers. Specifically, we compare the performance of CNNs trained with two types of galaxy images, stellar maps and dust-inclusive radiatively transferred images, each with three levels of observational realism: (1) no observational effects (idealized images), (2) realistic sky and point spread function (semirealistic images), and (3) insertion into a real sky image (fully realistic images). We find that networks trained on either idealized or semireal images have poor performance when applied to survey-realistic images. In contrast, networks trained on fully realistic images achieve 87.1 per cent classification performance. Importantly, the level of realism in the training images is much more important than whether the images included radiative transfer, or simply used the stellar maps ($87.1{{\ \rm per\ cent}}$ compared to $79.6{{\ \rm per\ cent}}$ accuracy, respectively). Therefore, one can avoid the large computational and storage cost of running radiative transfer with a relatively modest compromise in classification performance. Making photometry-based networks insensitive to colour incurs a very mild penalty to performance with survey-realistic data ($86.0{{\ \rm per\ cent}}$ with r-only compared to $87.1{{\ \rm per\ cent}}$ with gri). This result demonstrates that while colour can be exploited by colour-sensitive networks, it is not necessary to achieve high accuracy and so can be avoided if desired. We provide the public release of our statistical observational realism suite, RealSim, as a companion to this paper.
8 October 2020 in Monthly Notices of the Royal Astronomical Society
We investigate the spatial structure and evolution of star formation and the interstellar medium (ISM) in interacting galaxies. We use an extensive suite of parsec-scale galaxy merger simulations (stellar mass ratio = 2.5:1), which employs the ’Feedback In Realistic Environments-2’ model (fire-2). This framework resolves star formation, feedback processes, and the multi-phase structure of the ISM. We focus on the galaxy-pair stages of interaction. We find that close encounters substantially augment cool (HI) and cold-dense (H2) gas budgets, elevating the formation of new stars as a result. This enhancement is centrally-concentrated for the secondary galaxy, and more radially extended for the primary. This behaviour is weakly dependent on orbital geometry. We also find that galaxies with elevated global star formation rate (SFR) experience intense nuclear SFR enhancement, driven by high levels of either star formation efficiency (SFE) or available cold-dense gas fuel. Galaxies with suppressed global SFR also contain a nuclear cold-dense gas reservoir, but low SFE levels diminish SFR in the central region. Concretely, in the majority of cases, SFR-enhancement in the central kiloparsec is fuel-driven (55% for the secondary, 71% for the primary) - whilst central SFR-suppression is efficiency-driven (91% for the secondary, 97% for the primary). Our numerical predictions underscore the need of substantially larger, and/or merger-dedicated, spatially-resolved galaxy surveys - capable of examining vast and diverse samples of interacting systems - coupled with multi-wavelength campaigns aimed to capture their internal ISM structure.
3 February 2017 in Monthly Notices of the Royal Astronomical Society
The interpretive power of the newest generation of large-volume hydrodynamical simulations of galaxy formation rests upon their ability to reproduce the observed properties of galaxies. In this second paper in a series, we employ bulge+disc decompositions of realistic dust-free galaxy images from the Illustris simulation in a consistent comparison with galaxies from the Sloan Digital Sky Survey (SDSS). Examining the size–luminosity relations of each sample, we find that galaxies in Illustris are roughly twice as large and 0.7 mag brighter on average than galaxies in the SDSS. The trend of increasing slope and decreasing normalization of size–luminosity as a function of bulge fraction is qualitatively similar to observations. However, the size–luminosity relations of Illustris galaxies are quantitatively distinguished by higher normalizations and smaller slopes than for real galaxies. We show that this result is linked to a significant deficit of bulge-dominated galaxies in Illustris relative to the SDSS at stellar masses |$\log \mathrm{{\rm }M}_{\star }/\mathrm{M}_{{\odot }}\lesssim 11$|. We investigate this deficit by comparing bulge fraction estimates derived from photometry and internal kinematics. We show that photometric bulge fractions are systematically lower than the kinematic fractions at low masses, but with increasingly good agreement as the stellar mass increases.
30 August 2019 in Monthly Notices of the Royal Astronomical Society
We analyse the optical morphologies of galaxies in the IllustrisTNG simulation at z ∼ 0 with a convolutional neural network trained on visual morphologies in the Sloan Digital Sky Survey. We generate mock SDSS images of a mass complete sample of $\sim 12\, 000$ galaxies in the simulation using the radiative transfer code SKIRT and include PSF and noise to match the SDSS r-band properties. The images are then processed through the exact same neural network used to estimate SDSS morphologies to classify simulated galaxies in four morphological classes (E, S0/a, Sab, Scd). The CNN model classifies simulated galaxies in one of the four main classes with the same uncertainty as for observed galaxies. The mass–size relations of the simulated galaxies divided by morphological type also reproduce well the slope and the normalization of observed relations which confirms a reasonable diversity of optical morphologies in the TNG suite. However we find a weak correlation between optical morphology and Sersic index in the TNG suite as opposed to SDSS which might require further investigation. The stellar mass functions (SMFs) decomposed into different morphologies still show some discrepancies with observations especially at the high-mass end. We find an overabundance of late-type galaxies ($\sim 50{{\ \rm per\ cent}}$ versus $\sim 20{{\ \rm per\ cent}}$) at the high-mass end [log(M*/M⊙) > 11] of the SMF as compared to observations according to the CNN classifications and a lack of S0 galaxies ($\sim 20{{\ \rm per\ cent}}$ versus $\sim 40{{\ \rm per\ cent}}$) at intermediate masses. This work highlights the importance of detailed comparisons between observations and simulations in comparable conditions.
19 March 2021 in Monthly Notices of the Royal Astronomical Society
The Canada–France Imaging Survey (CFIS) will consist of deep, high-resolution r-band imaging over ∼5000 deg2 of the sky, representing a first-rate opportunity to identify recently merged galaxies. Because of the large number of galaxies in CFIS, we investigate the use of a convolutional neural network (CNN) for automated merger classification. Training samples of post-merger and isolated galaxy images are generated from the IllustrisTNG simulation processed with the observational realism code RealSim. The CNN’s overall classification accuracy is 88 per cent, remaining stable over a wide range of intrinsic and environmental parameters. We generate a mock galaxy survey from IllustrisTNG in order to explore the expected purity of post-merger samples identified by the CNN. Despite the CNN’s good performance in training, the intrinsic rarity of post-mergers leads to a sample that is only ∼6 per cent pure when the default decision threshold is used. We investigate trade-offs in purity and completeness with a variable decision threshold and find that we recover the statistical distribution of merger-induced star formation rate enhancements. Finally, the performance of the CNN is compared with both traditional automated methods and human classifiers. The CNN is shown to outperform Gini–M20 and asymmetry methods by an order of magnitude in post-merger sample purity on the mock survey data. Although the CNN outperforms the human classifiers on sample completeness, the purity of the post-merger sample identified by humans is frequently higher, indicating that a hybrid approach to classifications may be an effective solution to merger classifications in large surveys.