Context. Stellar bars are a common morphological feature of spiral galaxies. While it is known that they can form in isolation, or be induced tidally, few studies have explored the production of stellar bars in galaxy merging. We look to investigate bar formation in galaxy merging using methods from deep learning to analyse our N-body simulations. Aims. The primary aim is to determine the constraints on the mass ratio and orientations of merging galaxies that are most conducive to bar formation. We further aim to explore whether it is possible to classify simulated barred spiral galaxies based on the mechanism of their formation. We test the feasibility of this new classification schema with simulated galaxies. Methods. Using a set of 29 400 images obtained from our simulations, we first trained a convolutional neural network to distinguish between barred and non-barred galaxies. We then tested the network on simulations with different mass ratios and spin angles. We adapted the core neural network architecture for use with our additional aims. Results. We find that a strong inverse relationship exists between the mass ratio and the number of bars produced. We also identify two distinct phases in the bar formation process; (1) the initial, tidally induced formation pre-merger and (2) the destruction and/or regeneration of the bar during and after the merger. Conclusions. Mergers with low mass ratios and closely-aligned orientations are considerably more conducive to bar formation compared to equal-mass mergers. We demonstrate the flexibility of our deep learning approach by showing it is feasible to classify bars based on their formation mechanism.

Classifying the morphologies of galaxies is an important step in understanding their physical properties and evolutionary histories. The advent of large-scale surveys has hastened the need to develop techniques for automated morphological classification. We train and test several convolutional neural network architectures to classify the morphologies of galaxies in both a 3-class (elliptical, lenticular, spiral) and 4-class (+irregular/miscellaneous) schema with a dataset of 14034 visually-classified SDSS images. We develop a new CNN architecture that outperforms existing models in both 3 and 4-way classification, with overall classification accuracies of 83 per cent and 81 per cent respectively. We also compare the accuracies of 2-way / binary classifications between all four classes, showing that ellipticals and spirals are most easily distinguished (>98 per cent accuracy), while spirals and irregulars are hardest to differentiate (78 per cent accuracy). Through an analysis of all classified samples, we find tentative evidence that misclassifications are physically meaningful, with lenticulars misclassified as ellipticals tending to be more massive, among other trends. We further combine our binary CNN classifiers to perform a hierarchical classification of samples, obtaining comparable accuracies (81 per cent) to the direct 3-class CNN, but considerably worse accuracies in the 4-way case (65 per cent). As an additional verification, we apply our networks to a small sample of Galaxy Zoo images, obtaining accuracies of 92 per cent, 82 per cent and 77 per cent for the binary, 3-way and 4-way classifications respectively.

We study the morphologies of 3,964 galaxies and their progenitors with M⋆ > 1010M⊙ in the reference eagle hydrodynamical simulation from redshifts z = 1 to z = 0, concentrating on the redshift evolution of the bar fraction. We apply two convolutional neural networks (CNNs) to classify 35,082 synthetic g-band images across 10 snapshots in redshift. We identify galaxies as either barred or unbarred, while also classifying each sample into one of four morphological types: elliptical (E), lenticular (S0), spiral (Sp), and irregular/miscellaneous (IrrM). We find that the bar fraction is roughly constant between z = 0.0 to z = 0.5 (32 per cent to 33 per cent), before exhibiting a general decline to 26 per cent out to z = 1. The bar fraction is highest in spiral galaxies, from 49 per cent at z = 0 to 39 per cent at z = 1. The bar fraction in S0s is lower, ranging from 22 per cent to 18 per cent, with similar values for the miscellaneous category. Under 5 per cent of ellipticals were classified as barred. We find that the bar fraction is highest in low mass galaxies (M⋆ ≤ 1010.5M⊙). Through tracking the evolution of galaxies across each snapshot, we find that some barred galaxies undergo episodes of bar creation, destruction and regeneration, with a mean bar lifetime of 2.24 Gyr. We further find that incidences of bar destruction are more commonly linked to major merging, while minor merging and accretion is linked to both bar creation and destruction.