Observationally derived change in the star formation rate as mergers progress

TL;DR

This work develops a statistical framework to trace how star formation rate (SFR) changes as galaxy mergers progress by training a convolutional neural network on IllustrisTNG merger images with known merger times and applying it to KiDS observations. By comparing merging galaxies to ten closely matched non-merging controls (∆SFR = log(SFR_merger) − median(log(SFR_non-merger))), the study finds a pronounced SFR enhancement that rises from about 300 Myr before coalescence to at least 200 Myr after, peaking near the merger event and then gradually declining. The results show higher SFR enhancements for more massive galaxies, while the influence of local density is weaker and less certain due to merger-time uncertainties. The approach demonstrates the feasibility of deriving merger-time–resolved SFR trends from real data and provides insights consistent with simulation predictions about merger-driven star formation and its environmental and mass dependencies.

Abstract

Galaxy mergers can change the rate at which stars are formed. We can trace when these changes occur in simulations of galaxy mergers. However, for observed galaxies we do not know how the star formation rate (SFR) evolves along the merger sequence as it is difficult to probe the time before or after coalescence. We aim to derive how SFR changes in observed mergers throughout the merger sequence, from a statistical perspective. Merger times were estimated for observed galaxy mergers in the Kilo Degree Survey (KiDS) using a convolutional neural network (CNN). The CNN was trained on mock KiDS images created using IllustrisTNG data. The SFRs were derived from spectral energy density fitting to KiDS and VIKINGs data. To determine the change in SFR for the merging galaxies, each merging galaxy was matched and compared to ten comparable non-merging galaxies; matching each galaxy in redshift, stellar mass, and local density. Mergers see an increase in the SFR for galaxies from 300~Myr before the merger until coalescence, continuing until at least 200~Myr after the merger event. After this, there is a possibility that SFR activity in the mergers begins to decrease, but we need more data to better constrain our merger times and SFRs to confirm this. We find that more galaxies with higher stellar mass (M$_{\star}$) have greater SFR enhancement as they merge compared to lower-M$_{\star}$ galaxies. There is no clear trend of changing SFR enhancement as local density changes, but the least dense environments have the least SFR enhancement. The increasing SFR enhancement is likely due to the closer proximity of galaxies and the presence of more close passes as the time before the merger approaches 0~Myr, with the SFR slowing 200~Myr after the merger event.

Figures (14)

• Figure 1: Randomly selected example IllustrisTNG image after processing to appear like a KiDS image. The target galaxy is in the centre of the image. The left column shows the $u$, $g$, $r$, and $i$ band images, as labelled, with arcsinh scaling applied twice. The right column shows the associated segmentation maps for each band, with each colour representing a different segment. Each panel is $25.6 \times 25.6$ arcsec ($128 \times 128$ pixels), corresponding to a physical size of $53 \times 53$ kpc at the redshift of the IllustrisTNG galaxy ($z=0.11$).
• Figure 2: Randomly selected example KiDS image of a merging galaxy. The target galaxy is in the centre of the image. The left column shows the $u$, $g$, $r$, and $i$ band images, as labelled, with arcsinh scaling applied twice. The right column shows the associated segmentation maps for each band, with each colour representing a different segment. Each panel is $25.6 \times 25.6$ arcsec ($128 \times 128$ pixels), corresponding to a physical size of $44 \times 44$ kpc at the redshift of the galaxy ($z=0.09$).
• Figure 3: Predicted merger times of the test data against the true merger times. The colour corresponds to the number density from low (purple) to high (red). The red line indicates a one-to-one relation and the MSE for the normalised times is shown in the bottom right. Dark brown diamonds show the median true merger times and median predicted merger times in predicted merger time bins with widths of 200 Myr; x errors are the median absolute deviation of the true times in the bin and y errors are the median absolute deviation of the predicted times.
• Figure 4: Distribution of predicted merger times for KiDS merging galaxies. Most of the KiDS mergers are pre-mergers, with merger times less than 0 Myr.
• Figure 5: Sixteen KiDS merging galaxies with their associated predicted merger times (time after a merger). The shown images are $r$-band images with arcsinh scaling applied twice. The galaxy in the centre of the image is the galaxy whose merger time is shown. Each panel is $25.6 \times 25.6$ arcsec ($128 \times 128$ pixels), corresponding to a physical size of $53 \times 53$ kpc at the median redshift of the KiDS merger sample ($z=0.11$).