Confinement scaling with machine size in the updated ITPA global H-mode energy confinement database

A revision of the International Tokamak Physics Activity (ITPA) global H-mode energy confinement scaling in tokamaks was carried out in 2020, using the latest version of the multi-machine ITPA database (DB5.2.3), including data from machines with fully metallic walls (ASDEX Upgrade with the full tungsten wall and JET with the ITER-like wall). This resulted in the ITPA20 scaling law for ELMy H-modes, a power law estimated both in engineering and dimensionless form, exhibiting several dependencies that are different to those seen in the IPB98(y,2) scaling. One of the notable differences is a considerably weaker dependence on major radius. The present work aims to contribute to an explanation for this reduced size scaling. Although multicollinearity between the predictor variables may play a role, various multicollinearity diagnostics and mitigation techniques suggest that it is not the sole cause of the observed reduction. Optimisation and classification techniques reveal a subset of database points exhibiting weak size dependence, whereas the majority of points lie in a region characterized by strong size scaling, reminiscent of IPB98(y,2). Both clusters are relatively well separated in the space of normalised collisionality, gyroradius and pressure, with the standard operational point for ITER lying closest to the region of strong size scaling.