Abstract: Simulation of tokamak plasmas is a multi-physics challenge across multiple timescales. Coupled partial-differential-equations evolve quantities like current, heat, particles and momentum, over entire experiment timescales. Coefficients in these PDEs correspond to physics processes over much shorter timescales, like turbulent transport, collisions, MHD and plasma equilibrium. The dependence of these processes on the plasma state itself makes the system highly nonlinear. Even if each individual process is understood, non-trivial emergent phenomena can arise from the nonlinear interplay between these processes. Integrated modelling frameworks combine reduced physics models of these processes at various fidelity, and evolve the coupled PDE system.
Recent progress in the quality of reduced models has enabled accurate physics-based simulation over multiple confinement times. Such modelling allows predicting current, temperature, density and rotation profiles, disentangling the causality at play behind the modeled time evolution. Furthermore, advances in differentiable simulation and machine-learned physics model surrogates enable fast and accurate simulation for many-query use cases like pulse planning, optimization, and controller design. Physics gaps remain towards ITER operation and reactor design.
ITER-class devices will be opaque to neutrals and fueled by pellets, in absence of physics understanding of the transport in the pedestal, extrapolation is uncertain. In burning plasmas, core profiles are strongly non-linear coupled to the fusion power, making performance predictions relying on unverified reduced models.
This seminar we review the state-of-the-art in integrated modelling, highlighting examples of heightened physics understanding gained by high-fidelity integrated modelling, and progress in developing fast differentiable frameworks using ML-surrogates for optimization and control-oriented simulation.
From [W. Morris JPP colloquium Oct. 2024]
Go to the Events page.