“…Technically speaking, these first generation (G1) adaptive methods are biased, the bias being determined by the number of expansion coefficients selected in each variable. Because of these limitations, we developed another strategy that represents the RTE solution as a histogram over a mesh (G2) that is coarse initially but which is then refined based on an information density function constructed from both the forward and the adjoint RTE solutions (G3) [13, 16, 17]. Although the initial mesh is revised in an automated way using this strategy, it is still necessary to define an initial mesh decomposition, which makes this option problem-dependent as well.…”