The most widely accepted unified scaling law for the energy confinement time in stellarators is a power law that contains a normalization factor for each individual device (and even for each sufficiently different magnetic configuration in a single machine). In the last decade, new and very powerful data analysis tools, based on symbolic regression (SR) via genetic programming (GP), have become quite consolidated and have provided very interesting results for tokamak configurations. The first application of SR via GP to the largest available multimachine stellarator database permits us to relax the power law constraint as an alternative to the use of renormalization factors. This approach, implemented with well-understood model selection criteria for the fitness function based on information theory and Bayesian statistics, has allowed convergence on very competitive global scaling laws, which present exponential terms, but do not contain any renormalization coefficient. Moreover, the exploratory application of SR via GP has revealed that the two main types of magnetic topology, those with and without shear, can be much better interpreted using two different models. The fact that these new scaling laws have been derived without recourse to any renormalization increases their interpretative value and confirms the dominant role of turbulence in determining the confinement properties of the stellarator. On the other hand, the techniques developed emphasise the need to improve the statistical basis before drawing definitive conclusions and providing reliable extrapolations.
A new approach to the formulation and validation of scaling expressions for plasma confinement in tokamaks