Investigation of density limit parameter dependence using symbolic regression and MHD simulation in EAST

Abstract The density limit is a fundamental operational boundary in tokamaks. The accurate prediction of this limit is crucial for attaining higher-density, higher-confinement plasma operations while mitigating the risk of disruptions. Although the Greenwald scaling law is a classical formula used to predict this density limit, recent experiments on the EAST tokamak have repeatedly surpassed its predictions. Therefore, there is an urgent need to establish a more accurate predictive formula for the density limit on EAST. To address this challenge, this work first defines the density limit operationally by the onset of a key physical precursor: the multifaceted asymmetric radiation from the edge (MARFE). Building on this criterion, we apply symbolic regression (SR) with genetic programming, constrained by physics-informed priors, to derive a compact and interpretable density limit formula for EAST directly from a finite set of experimental data. On the unseen test set, the density limit formula achieves (mean absolute percentage error) MAPE = 7.33%. Compared with Greenwald’s results on the test set (MAPE = 32.59%), our formula offers higher predictive accuracy. By performing magnetohydrodynamic simulations with parameter scans over scrape-off layer power, q 95 , elongation, and minor radius, we obtained their scaling exponents whose scaling relationships exhibit a trend consistent with the SR results. Further simulations indicate that enhanced thermal cooling leads to increased MARFE displacement, while thermal contraction further drives the growth of tearing mode. The subsequent development of tearing mode may eventually trigger disruptions, which could explain why density-limit disruptions in EAST often occur following MARFE movement.