Computations in fusion physics (Q1102828)

The modern approach to nuclear fusion uses inertial confinement by a high-power laser beam, and the computer codes used to simulate the plasma behaviour now contain a lot of physics. Nonlinear interaction of a laser with a turbulent plasma is a major topic of experimental and numerical study. Simulation can be on either the particle or the hydrodynamic scale. A moving plasma generates an EM field which is computed self- consistently in time steps small enough to resolve the shortest period, in particle codes typically that of the electron plasma oscillations. A grid spacing of the order of the Debye length suffices, which is equivalent to considering the particles to have a (large) finite size. Linear interpolation of particle charge into neighbouring cells is correct to dipole order, and the electric field can be computed by a direct finite difference solution of Poisson's equation or by Fourier Transform and linearly interpolated to the particle positions. With care, the 3 types of numerical instability - beaming, aliasing and time- integration - can be nearly eliminated. Laser-plasma interaction requires inclusion of magnetic and relativistic effects, and the plasma can develop bubbles and ripples due to nonlinearity. The regions and times that can be simulated are much smaller than those used in experiments. At the extremely high densities required for fusion the plasma becomes more like a solid and is strongly coupled; the Coulomb energy exceeds the kinetic energy and the electron levels are filled up to the Fermi energy. Determination of thermodynamic and transport properties from the radial distribution function is simplified by using a ``one-component plasma'' model, and various numerical methods have been employed. Some hydrodynamic codes include as much physics as possible at the expense of sophisticated numerical methods. They led to scaling laws a decade ago, and are used to design and interpret experiments such as to determine heat flow through the plasma, which is found to be less than expected. Most codes have a user-friendly front-end for problem description and a back-end for data analysis. The physics involves widely differing time and space scales, which complicates the design of the algorithm. This cycles through the physical processes involved: numerical methods for ionization, hydrodynamic, thermonuclear reactions and thermal transport are discussed. The Incomplete Cholesky Conjugate Gradient method is shown to be very effective for solving the discretized thermal diffusion equation. Codes are used to explore the parameter space available to an experiment, perform numerical experiments with non- physical parameter values, and extrapolate and interpolate experimental results. Future codes must be improved to keep track of improving experiments, and will need to use parallel computing.