The effect of galaxy interactions on star formation rates in the COLIBRE simulations

The paper investigates the effect of galaxy interactions on star formation rates (SFRs) in the COLIBRE simulations, a set of cosmological hydrodynamical simulations of galaxy formation. The study focuses on star-forming galaxies at redshift z about 0, using the largest COLIBRE simulation at m6 resolution (gas particle mass m gas = 1.8 times 10 6, M) with a volume of (200, cMpc) cubed. The authors construct samples of interacting galaxies (with closest companions having stellar masses > 10% of the interacting galaxy's stellar mass) and matched isolated control galaxies, matched on stellar mass, redshift, local density, and isolation, following the method of Patton et al. (2016).

Key findings on the interaction-induced specific star formation rate (sSFR) enhancement:

1. Dependence on pair separation and stellar mass: The mean sSFR of interacting galaxies with stellar masses 10 10 < M*, int/M < 10 12 is enhanced relative to controls for pair separations r 3D 200, kpc. The enhancement increases monotonically with decreasing separation, reaching Q(sSFR) about 1.7 in the smallest separation bin (r 3D = 5, kpc). At separations r 3D 1.2 for galaxies with M*, int 10 10.5, M, peaking at Q about 1.6 for M*, int about 10 9, M. The enhancement weakens for more massive galaxies, approaching Q about 1 by M*, int about 10 11, M.

2. Dependence on mass ratio: The sSFR enhancement is stronger for higher mass ratios (closer to unity). For mass ratio bins 1/10 < M*, cc/M*, int < 1/3, 1/3 < M*, cc/M*, int < 3, and 3 < M*, cc/M*, int < 10, the enhancement peaks at Q about 1.5, 1.6, and 1.7, respectively, around M*, int about 10 9, M. At higher masses (10 9.5 < M*, int/M < 10 11), the enhancement remains visible for low and intermediate mass ratios but decreases to unity for the highest mass ratio bin.

3. Aperture dependence: The measured sSFR enhancement is strongly dependent on the aperture within which sSFR is computed. For the smallest separation bin (r 3D < 10, kpc), the enhancement is Q about 1.7 for a 10 kpc aperture (fiducial), about 1.95 for a 3 kpc aperture, about 2.15 for the stellar half-mass radius (R 1/2), and about 2.6 for a 1 kpc aperture.

4. Gas fractions: The molecular gas fraction enhancement peaks at M*, int about 10 9, M, with a shape and amplitude similar to the sSFR enhancement. The atomic gas fraction is enhanced for galaxies with M*, int 10 9.5, M, with the enhancement rising monotonically with decreasing stellar mass.

5. Contribution to cosmic SFR density: The contribution of pre-merger galaxy interactions (for interacting galaxies with stellar masses 10 8 < M*, int/M < 10 12, mass ratios 0.1 < M*, cc/M*, int < 10, and across all separations) to the cosmic SFR density at z about 0 is approximately 2.1 per cent.

6. Comparison with SDSS observations: The authors compare COLIBRE predictions with SDSS data from Patton et al. (2013), using projected separations, line-of-sight velocity cuts, and statistical controls. They find that COLIBRE reproduces the shape of the mean sSFR enhancement as a function of projected separation, with both showing enhancement out to r proj about 150, kpc. However, COLIBRE underpredicts the normalization by a factor of approximately 2, with the largest difference in the lowest separation bin (Q about 1.6 in COLIBRE vs. Q about 2.65 in SDSS at r proj = 5, kpc).

7. Resolution and volume convergence: The sSFR enhancement is converged with simulation volume but increases with numerical resolution. At m7 resolution, the enhancement is lower (e.g., Q about 1.45 vs. about 1.7 at m6 in the lowest separation bin) and extends only out to r 3D < 130, kpc (vs. < 200, kpc at m6). The authors suggest that higher resolution (m5) could reduce the discrepancy with SDSS.

8. Verification of interaction-driven enhancement: By tracing the evolution of interacting galaxies and controls back in time, the authors show that the sSFR differences between interacting and control galaxies disappear at z about 0.3, when the median separation grows beyond about 200, kpc, confirming that the enhancement is due to interactions rather than environmental differences.

The paper concludes that galaxy interactions enhance SFRs in COLIBRE, with the enhancement depending on separation, stellar mass, mass ratio, and aperture size, and that the predicted enhancement is qualitatively consistent with observations but lower in normalization.

Improvements to AI Systems Based on This Paper:

1. Aperture-Aware Star Formation Rate Prediction Models

• Improvement: Train AI models to predict sSFR enhancement as a function of aperture size (e.g., 1 kpc, 3 kpc, 10 kpc, half-mass radius) rather than a single fixed aperture.

• What the improved system can do: Automatically correct for aperture bias in galaxy surveys, enabling fair comparisons between simulations and observations with different spatial resolutions.

2. Mass-Ratio and Separation-Dependent Interaction Classifiers

• Improvement: Build a neural network that classifies galaxy pairs by interaction stage (pre-merger, merging, post-merger) using stellar mass ratio and 3D separation, trained on COLIBRE’s enhancement curves.

• What the improved system can do: Predict the expected sSFR boost for any given pair configuration, allowing rapid identification of galaxies most likely to experience starbursts in large photometric or spectroscopic catalogs.

3. Resolution-Compensated Simulation-to-Observation Mapper

• Improvement: Develop a domain-adaptation model that learns the systematic offset in sSFR enhancement between simulation resolutions (m6 vs. m7) and applies a correction to match observed SDSS normalization.

• What the improved system can do: Produce observationally calibrated predictions from lower-resolution simulations, saving computational cost while maintaining accuracy for galaxy evolution studies.

4. Time-Dependent Interaction Feedback Loop for Cosmological Simulations

• Improvement: Implement a reinforcement learning agent that adjusts subgrid star formation feedback parameters in real-time based on predicted interaction-induced enhancement (from the paper’s findings), rather than using fixed values.

• What the improved system can do: Self-consistently model how interactions alter gas fractions and SFRs over cosmic time, improving the fidelity of future hydrodynamical simulations without manual tuning.

5. Cosmic SFR Density Contribution Estimator

• Improvement: Create a generative model that, given a galaxy population’s mass function and pair statistics, predicts the fractional contribution of pre-merger interactions to the total cosmic SFR density (calibrated to the 2.1% figure).

• What the improved system can do: Quickly estimate interaction-driven star formation in any simulated or observed volume, enabling real-time forecasts for survey design (e.g., LSST, Euclid) and theoretical comparisons.

6. Contrastive Learning for Isolated vs. Interacting Galaxy Embeddings

• Improvement: Use the matched control sample methodology (Patton et al. 2016) to train a contrastive encoder that separates interacting from isolated galaxies in feature space, using stellar mass, local density, and separation as conditioning variables.

• What the improved system can do: Automatically detect subtle interaction signatures in multi-wavelength data (e.g., gas fraction, morphology) without requiring explicit pair catalogs, improving outlier detection in galaxy surveys.

7. Predictive Model for Gas Fraction Enhancement

• Improvement: Train a regression model that predicts molecular and atomic gas fraction enhancements as functions of stellar mass and pair separation, using the paper’s findings (peaks at 10 9 M, monotonic rise for atomic gas at low mass).

• What the improved system can do: Provide physically motivated priors for radio/sub-mm observations (e.g., ALMA, SKA) to infer gas content in unresolved interacting galaxies, aiding in the interpretation of faint signals.