Theoretical and numerical analysis of a minimal residual solver for 2D Boltzmann transport equation (Q1863296)

Let \(D\subset\mathbb{R}^2\) be a bounded open set with Lipschitz boundary \(\partial D\). Suppose that the outer normal \({\mathbf n}(x)\) to \(\partial D\) exists almost everywhere. Let us introduce the following notation: \(\mathcal Q=D\times B\), \(\Gamma^-:=\{(x,\Omega)\subset\partial D\times B\), \(\Omega\cdot{\mathbf n}(x) < 0\}\), where \(B = \{\Omega\in\mathbb{R}^2\), \(\|\Omega\|_2 < 1\}\). The authors consider the problem of finding the solution \(f :\mathcal Q\to\mathbb{R}\) of Boltzmann transport equation in dimension two: \[ \left\{\begin{aligned} \Omega \cdot \nabla_xf(x, \Omega) + \sigma(x)f(x,\Omega)&= \int_Bk(x, \Omega,\Omega')f(x,\Omega') d\Omega'+ S(x,\Omega) \text{in} \mathcal Q,\\ f(x,\Omega) &= 0\quad \text{on}\quad \Gamma^-. \end{aligned}\right.\tag{1} \] The function \(f(x,\Omega)\) describes the behavior of neutrons in \(D\), in interaction with them, representing, up to some factor, the flux of neutron density at the position \(x\) with velocity \(\Omega\subset B \). The function \(\sigma(x)\) accounts for neutron-domain interaction, whereas a positive kernel \(k(x, \Omega, \Omega')\) describes collisions between neutrons. At last, the neutron source is represented by a nonnegative function \(S(x,\Omega)\). Relying on the splitting of the collision operator, the authors prove theoretical convergence for an infinite dimensional adaptation of the minimal residual algorithm for the integro-differential equation (1). The algorithm is numerically compared with some iterative methods developed in the last few years.