Many proposed models of physics beyond the Standard Model of particle physics (BSM) predict new resonances. Thus, a single experimental search for such new resonances can probe a multitude of BSM models. In particular, searches for resonances decaying into hadrons can probe for any resonances that can be produced at the LHC. However, such dijet searches lose sensitivity at low masses due to transverse momentum trigger thresholds.

This thesis presents a search targeting events where the BSM resonance is produced back-to-back with a photon radiated from the initial state partons. The search achieves sensitivity to BSM resonance masses as low as 20 GeV for the first time using data from the ATLAS detector. This is done by triggering on the initial state radiation (ISR), which decouples the accessible resonance mass from the trigger thresholds. Trigger thresholds instead imply a minimum boost of the resonance, causing the lowest-mass resonances to produce overlapping jets that are best reconstructed as a single large-radius jet.

Precise large-radius jet reconstruction down to very low masses is thus a crucial requirement of the search. However, the standard Run 2 large-radius jet definition is only calibrated down to 50 GeV. To overcome this limitation, in this thesis the resonance candidate jets are instead reconstructed by reclustering calibrated 𝑅 = 0.2 hadronic anti−𝑘_T jets and inner detector tracks into so-called track-assisted reclustered (TAR) jets.

A dataset corresponding to 140.1 fb⁻¹ of proton-proton collisions at √𝑠 = 13 TeV recorded with the ATLAS detector from 2015 through 2018 is analysed. The most significant excess over the SM expectation is localized near 40 GeV of jet mass, but at 2.8 𝜎 it is not significant. Thus, upper limits are set on the cross-section for producing 𝑍′ vector resonances. The 95% CL𝑠 limits range from 0.02 pb for a 20 GeV resonance to 0.2 pb for a 225 GeV resonance.