Leveraging Bayesian optimization for automated plasma composition optimization of ARC with physics-based turbulent transport models

New tokamaks are usually designed using empirical scaling laws. However, these cannot capture some of the impacts of changing the plasma composition via impurities. Here, we optimize the plasma composition of ARC (Howard, et al., JPP, Submitted 2025). Modeling is performed using MAESTRO, an integrated modeling tool using TGLF as the transport model (Staebler, NF, 2021) and EPED for pedestal predictions (Snyder, NF, 2011). A relatively high effective charge (Zeff) and low main ion fraction operating point is identified to maximize fusion power. Changing the main ion fraction and Zeff result in a competition between increased temperature peaking and reduced densities. Bayesian optimization is employed to expedite the process of finding the best operating point, allowing exploration of pedestal density and geometric shaping parameters. The uncertainties of MAESTRO due to TGLF saturation rule variation, EPED uncertainty, and seeding variation are quantified. Ongoing work includes establishing a success criterion, which enforces access to H-mode and core-edge compatibility.