A Study of the Substructure of Jets Initiated by Top Quarks and W Bosons

Large hadronic jets initiated by the decays of heavy objects such as top quarks and $W$ bosons have long been a very active field of study in particle physics. The Lund Jet Plane is a new and versatile tool for investigating the sub-structure of these complex objects. This thesis presents a collection of studies involving this new observable including contributions to its first-ever measurement in $p-p$ collision data. A brand new measurement of the Lund Jet Plane is also presented, as well as the development of new tagging techniques involving Deep Learning technologies. All measurements use the ATLAS data collected during Run 2, from 2015 to 2018, at $\sqrt{s} = 13 \mathrm{TeV}$. A brand new measurement of the Lund Jet Plane in semi-leptonic $t\bar{t}$ decays is presented. It is the first measurement of this observable for boosted-object topologies using $R=1.0$ jets. The measurement complements other measurements of the LJP which have only probed small-radius jets predominantly initiated by light quarks and highly energetic gluons. The measurement is unfolded to the particle level using Iterative Bayesian Unfolding. The unfolded measurement is compared to a wide range of Monte Carlo event generators. Measurements of new jet substructure observables, like the Lund Jet Plane, in new jet topologies are valuable for understanding the composition of these jets and can help improve modelling by providing new tunes for Monte Carlo generators. The performance of various $W$ and top jet taggers is evaluated and calibrated using the data to provide new tagging recommendations for ATLAS analyses. A completely new class of $W$ and top jet taggers is also introduced. These taggers make use the many possible representations of the Lund Jet Plane as input to state-of-the art Deep Neural Network architectures, including Deep Convolutional Neural Networks, Recurrent Neural Networks and Graph Neural Networks.