Yong Sheng Koay

In physics, Lagrangians provide a systematic way to describe laws governing physical systems. In the context of particle physics, they encode the interactions and behavior of the fundamental building blocks of our universe. By treating Lagrangians as complex, rule-based constructs similar to linguistic expressions, we trained a transformer model — proven to be effective in natural language tasks — to predict the Lagrangian corresponding to a given list of particles. We report on the transformer’s performance in constructing Lagrangians respecting the Standard Model \mathrm{SU}(3)× \mathrm{SU}(2)× \mathrm{U}(1) S U ( 3 ) × S U ( 2 ) × U ( 1 ) gauge symmetries. The resulting model is shown to achieve high accuracies (over 90%) with Lagrangians up to six matter fields, with the capacity to generalize beyond the training distribution, albeit within architectural constraints. We show through an analysis of input embeddings that the model has internalized concepts such as group representations and conjugation operations as it learned to generate Lagrangians. We make the model and training datasets available to the community. An interactive demonstration can be found at: https://huggingface.co/spaces/JoseEliel/generate-lagrangians.

We perform a detailed study of the current phenomenological status of baryon number violating operators within the framework of the R-parity violating Minimal Supersymmetric Standard Model (RPV-MSSM). This study aims to identify any gaps in the experimental coverage of the RPV landscape. We identify the unique final states for all possible LSPs decaying via four different benchmark UDD operators. Both the direct production of the LSP and its production via gauge-cascades are considered. For each LSP, we assume that only one UDD coupling is non-zero at a time and confront the signals with existing ATLAS and CMS searches implemented in the recasting framework CheckMATE 2. We find that the UDD colored LSP sector is well covered with the mass bounds on the gluino LSP being the strongest, and with possible improvements for some of the right-handed squark LSPs. We also point out that there is limited coverage for electroweakino and slepton LSPs with UDD decays. This limitation may be due to the lack of targeted experimental searches for these specific final states or the appropriate recasting of existing searches.

We perform a classification of all potential supersymmetric $R$-parity violating signatures at the LHC to address the question: are existing bounds on supersymmetric models robust, or are there still signatures not covered by existing searches, allowing LHC-scale supersymmetry to be hiding? We analyze all possible scenarios with one dominant RPV trilinear coupling at a time, allowing for arbitrary LSPs and mass spectra. We consider direct production of the LSP, as well as production via gauge-cascades, and find 6 different experimental signatures for the $LL\bar E$-case, 6 for the $LQ\bar D$-case, and 5 for the $\bar U\bar D\bar D$-case; together these provide complete coverage of the RPV-MSSM landscape. This set of signatures is confronted with the existing searches by \texttt{ATLAS} and \texttt{CMS}. We find all signatures have been covered at the LHC, although not at the sensitivity level needed to probe the direct production of all LSP types. For the case of a dominant $LL\bar E$-operator, we use \texttt{CheckMATE} to quantify the current lower bounds on the supersymmetric masses and find the limits to be comparable to or better than the $R$-parity conserving case. Our treatment can be easily extended to scenarios with more than one non-zero RPV coupling.