To investigate errors caused by angular differencing in approximating the streaming terms of the transport equation, five different approximations are evaluated for three test problems in one-dimensional spherical geometry. The following schemes are compared: diamond, special truncation error minimizing weighted diamond, linear continuous (the original SN scheme), linear discontinuous, and new quadratic continuous. To isolate errors caused by angular differencing, the approximations are derived from the transport equation without spatial differencing, and the resulting coupled ordinary differential equations (ODEs) are solved with an ODE solver. Results from the approximations are compared with analytic solutions derived for two-region purely absorbing spheres. Most of the approximations are derived by taking moments of the conservation form of the transport equation. The quadratic continuous approximation is derived taking the zeroth moment of both the transport equation and the first angular derivative of the transport equation. The advantages of this approach are described. In all of the approximations, the desirability is shown of using an initializing computation of the = -1 angular flux to correctly compute the central flux and of having a difference approximation that ensures this central flux is the same for all directions. The behavior of the standard discrete ordinates equations in the diffusion limit is reviewed, and the linear and quadratic continuous approximations are shown to have the correct diffusion limit if an equal interval discrete quadrature is used.

In all three test problems, the weighted diamond difference approximation has smaller maximum and average relative flux errors than the diamond or the linear continuous difference approximations. The quadratic continuous approximation and the linear discontinuous approximation are both more accurate than the other approximations, and the quadratic continuous approximation has a decided edge over the linear discontinuous approximation in relative flux errors. The diamond, weighted diamond, and linear continuous approximations show quadratic system absorption and system leakage error reduction behavior with increasing N. The linear discontinuous and quadratic continuous approximations show fourth-order error reduction in these quantities. In one of the two-region test problems, the slope of the exact angular flux changes from nearly vertical to nearly horizontal at those points in the exterior region at which the interior region source just becomes visible. At these spatial points, errors in the continuous approximations propagate to each successive outgoing direction, leading to an oscillatory spatial error. The discontinuous approximation does not propagate these errors, although errors near the point of rapid slope change are larger than in the quadratic continuous approximation.