Fourier analysis of diamond discretization in particle transport (Q2569819)

This is a meticulous work on the so-called diamond discretisation (DD) method, an explicit second--order accurate finite difference scheme. Such scheme turns out to be frequently applied to linear dissipative first--order scalar hyperbolic equations that arise from the velocity discretisation of the free particle full transport equations. As these equations model, among other phenomena, nuclear particle propagation and the energy density of high frequency waves in scattering media, they turn out to be an important tool to study nuclear reactors, radiotherapy and electromagnetic waves as well as earthquake--generated shear waves. The standard Fourier transform technique is employed to analyse how DD performs for the linear first-order hyperbolic transport problem \[ \begin{cases} \beta \cdot \nabla u + \gamma u\;=\;0 \;& \;\text{ in } (0,L)\times(0,1)\; \\ u\;= \;g& \;\text{ on }(0,L)\times\{0\}\\ \end{cases}\tag{1} \] (In (1), the function \(g\), the constant \(L>0\), the unit vector \(\beta\) and the real \(\;\gamma\;\) are all given.) It is proved that DD behaves quite well for smooth boundary data and a small absorption coefficient. Being an \(L^2\)--isometry in the conservative case, it fails to damp high frequencies, just as happens with any other finite difference method. This leads to the presence of unphysical modes in the dissipative case, for nonsmooth boundary conditions. The thorough theoretical analysis, backed by numerical simulations, yields a better understanding of the whole range of spuriously generated oscillations. Computer experiments also confirm that the obtained theoretical estimates are optimal. Although slow \(L^2\)--norm convergence may occur for DD generated solutions when nonsmooth boundary conditions are present, suitable moments of these solutions still display an accuracy of order \(\;h^2\;\). (Observe that the DD mesh is built with squares, so that only one discretization parameter shows up.) Thus the application domain of DD depends on the information sought -- \(\;L^2\) norms or spatial moments. The presented results validate DD as a dependable tool for computing particle distributions in diffusive media, where absorption is weak and propagation along straight lines short. They also show that DD should be used with care when dealing with long distance particle propagation.