The design of fusion devices is typically based on computationally expensive simulations. This can be alleviated using high aspect ratio models that employ a reduced number of free parameters, especially in the case of stellarator optimization where non-axisymmetric magnetic fields with a large parameter space are optimized to satisfy certain performance criteria. However, optimization is still required to find configurations with properties such as low elongation, high rotational transform, finite beta and good fast particle confinement. In this work, we train a machine learning model to construct configurations with favourable confinement properties by finding a solution to the inverse design problem, that is, obtaining a set of model input parameters for given desired properties. Since the solution of the inverse problem is non-unique, a probabilistic approach, based on mixture density networks, is used. It is shown that optimized configurations can be generated reliably using this method.

Motivated by explosive releases of energy in fusion, space and astrophysical plasmas, we consider the nonlinear stability of stratified magnetohydrodynamic equilibria against two-dimensional interchanges of straight magnetic-flux tubes. We demonstrate that, even within this restricted class of dynamics, the linear stability of an equilibrium does not guarantee its nonlinear stability: equilibria can be metastable. We show that the minimum-energy state accessible to a metastable equilibrium under non-diffusive two-dimensional dynamics can be found by solving a combinatorial optimisation problem. These minimum-energy states are, to good approximation, the final states reached by our simulations of destabilised metastable equilibria for which turbulent mixing is suppressed by viscosity. To predict the result of fully turbulent relaxation, we construct a statistical mechanical theory based on the maximisation of Boltzmann's mixing entropy. This theory is analogous to the Lynden-Bell statistical mechanics of collisionless stellar systems and plasma, and to the Robert–Sommeria–Miller theory of two-dimensional vortex turbulence. Our theory reproduces well the results of our numerical simulations for sufficiently large perturbations to the metastable equilibrium.

The heat conductivity of a plasma is usually much higher along the magnetic field than across it, and, as a result, the presence of a magnetic island can significantly affect the temperature profile in its vicinity. Radiation energy losses, which depend sensitively on temperature, are thus strongly affected by magnetic islands. This phenomenon is explored in a simple mathematical setting, and it is shown that the presence of a magnetic island greatly enhances a plasma's capacity to radiate energy. In the limit of highly anisotropic heat conductivity, the steady-state heat conduction equation can be reduced to an ordinary differential equation. Although this equation operates in one dimension, the topology is not that of the real line, but corresponds to a rod with a cooling fin. As parameters such as the incoming heat flux or the radiation amplitude are varied, the radiation has a tendency to linger around the island, in particular in the region of the separatrix, and the total radiated energy is then significantly increased. The island acts as a ‘cooling fin’ to the plasma. Furthermore, the solutions exhibit bifurcations, where the location of the radiation zone suddenly changes.

Electron cyclotron resonance ion thrusters (ECRITs) have the potential to be used for space gravitational wave detection due to their wide thrust range. However, an unclear understanding of dynamic processes of ECRITs with strongly coupled multi-operating parameters limits further improvements on thrust noise and response velocity by feedback control systems. An integrative mathematical model considering the non-Maxwell electron energy distribution function for ECRITs is validated by experiments and used to study the influence of operating parameters on the dynamic processes of thrusters, which provides a new simplified grid model. Simulation results show the response processes with microwave (MW) power can be divided into two stages. The characteristic times of the first and second stages are respectively several microseconds and 10 ms, which are respectively dominated by plasma motion and the volume effect. The overshoot of screen grid (SG) current decreases, while its response time remains unchanged when the response time of MW power is prolonged. The response time of SG current with a step increase of flow rate is approximately 10 ms, consistent with the volume effect. The SG current decreases with rise of flow rate for high flow rate operations due to the small increment of ion density limited by low electron temperature, the decrease of ion Bohm velocity and reduction of sheath extraction area. The influence of grid voltage on the dynamic process of the SG current depends on variation ranges of extraction capabilities. When variations of sheath extraction area are limited, the response time is 5 μs, consistent with plasma response time. It is prolonged to 0.5 ms if sheath extraction area variations are large because they cause obvious variations of plasma parameters in the discharge chamber. These dynamic results can not only facilitate designing feedback controllers of micro-propulsion systems for high-precision space missions, but also provide guidance for ion sources to generate highly stable or rapid-response ion beam.

Astrobiology is often defined as the study of the origin, evolution, distribution and future of life on Earth and in the Universe and thought of as a discipline. In practice though, the delineation of astrobiology-related research and corresponding groups of researchers is far from straightforward. Here, we propose to apply text-mining methods to identify researcher communities depending on thematic similarities in their published works. After fitting a latent Dirichlet allocation topic model to the complete article corpus of three flagship journals in the field – Origins of Life and Evolution of Biospheres (1968–2020), Astrobiology (2001–2020), the International Journal of Astrobiology (2002–2020) – and computing author topic profiles, researcher communities are inferred from topic similarity networks to which community detection is applied. Such semantic social networks reveal, as we call them, ‘hidden communities of interest’ that gather researchers who publish on similar topics. The evolution of these communities is also mapped through time, bringing to light the significant shifts that the discipline underwent in the past 50 years.