Determining the Time Before or after a Galaxy Merger Event

Aims. This work aims to reproduce the time before or after a merger event ofmerging galaxies from the IllustrisTNG cosmological simulation using machinelearning. Methods. Images of merging galaxies were created in the u, g, r, and i bandsfrom IllustrisTNG. The merger times were determined using the time differencebetween the last simulation snapshot where the merging galaxies were tracked astwo galaxies and the first snapshot where the merging galaxies were tracked asa single galaxy. This time was then further refined using simple gravitysimulations. These data were then used to train a residual network (ResNet50),a Swin Transformer (Swin), a convolutional neural network (CNN), and anautoencoder (using a single latent neuron) to reproduce the merger time. Thefull latent space of the autoencoder was also studied to see if it reproducesthe merger time better than the other methods. This was done by reducing thelatent space dimensions using Isomap, linear discriminant analysis (LDA),neighbourhood components analysis, sparse random projection, truncated singularvalue decomposition and uniform manifold approximation and projection. Results. The CNN is the best of all the neural networks. The performance ofthe autoencoder was close to the CNN, with Swin close behind the autoencoder.ResNet50 performed the worst. The LDA dimensionality reduction performed thebest of the six methods used. The exploration of the full latent space producedworse results than the single latent neuron of the autoencoder. For the testdata set, we found a median error of 190 Myr, comparable to the time separationbetween snapshots in IllustrisTNG. Galaxies more than ≈ 625 Myr beforea merger have poorly recovered merger times, as well as galaxies more than≈ 125 Myr after a merger event.