Solving two-point boundary value problems by means of deficient quartic splines (Q1805262)

The lacunary interpolation problem considered is as follows: let \(x_k = kh\) for \(k = 0(1)n\) and \(h = 1/n\), find a quartic spline \(s(x)\) of deficiency 2 which interpolates the data \(s(x_k) = y_k\) and \(s''(x_k) = y_k{''}\) for \(k = 1(1)n-1\) along with boundary conditions \(s(x_0) = y_0\), \(s(x_n) = y_n\), or more generally, \({\mathcal B}_i (s(x_0), s'(x_0), s(x_n) s'(x_n)) = 0\) for \(i = 1,2\) where \(s(x) = s_k(x)\) for \(x \in [x_k, x_{k + 1}]\), and \(k = 0(1) n - 1\). The application to two-point boundary value problems involves interpolating the numerical solution of \(y'' = f(x, y, y')\) with boundary conditions \({\mathcal B}_i (y(0), y'(0), y(1), y'(1)) = 0\) for \(i = 1,2\), which has been obtained via, e.g., finite differences or shooting methods. For reasons of symmetry, \(n\) is taken to be odd and \(j := (n - 1)/2\) and \(y_m := y''(1/2)\). Writing \(s_0(x) =\) \(a_{0,0} + (x - x_0) a_{0,1} + (x - x_0)^2 y_0{''}/2! + (x - x_0)^3 a_{0,3} /3! + (x - x_0)^4 a_{0,4} /4!\) and for \(k = 1(1)n - 1\), \(s_k(x) = y_k + (x - x_0) a_{k,1} + (x - x_0)^2 y_0{''}/ 2! + (x - x_0)^3 a_{k,3} /3! + (x - x_0)^4 a_{k,4} /4!\), the authors show how the smoothness conditions yield a system which can be solved in \({\mathcal O} (n)\) flops for the \(a_{k,j}\) for \(k = 1(1)n - 1\) and reasonable starting values \(a_{0,0}^{(0)}\), \(a^{(0)}_{n - 1,3}\) for Newton's method for solving the equations \({\mathcal B}_i (a_{0,0}, a_{n - 1,3}) = 0\), \(i = 1,2\), which arise from the boundary conditions. A fourth-order convergence result is shown: If \(y \in {\mathcal C}^5 [0,1]\), then \(|s_k - y|_\infty = {\mathcal O} (h^4)\).