A new decoupling technique for the Hermite cubic collocation equations arising from boundary value problems (Q1840901): Difference between revisions

For the second order linear ordinary differential equation \[ L[u]\equiv u''(x)+a(x)u'(x)+b(x)u(x)=f(x) \] defined on a finite interval, Dirichlet et Neumann boundary value problems are considered. To these problems the method of Hermite collocation is applied. To this end the interval of definition of the equation is divided into \(N\) subintervals of equal lenth by the points \(x_0,x_1,\cdots,x_N\). On each subinterval \([x_{k-1},x_{k+1}]\) two standard basic cubic polynomials \(p_k(x)\) and \(q_k(x)\) satisfying \(p_k(x_j)=\delta_{k,j}\), \(p'_k(x_j)=0\), \(q_k(x_j)=0\), \(q'_k(x_j)=\delta_{k,j}\), for \(k,j=k-1,k,k+1\) are defined. Moreover on each subinterval \([x_k,x_{k+1}]\) two collocation points are choosen \(\tau_{2k+1}={1\over 2}(x_k+x_{k+1})-\theta h\) and \(\tau_{2k+1}={1\over 2}(x_k+x_{k+1})+\theta h\), where \(h=x_{k+1}-x_k\) and \(0<\theta<{1\over 2}\). The coefficients \(P_k\) and \(Q_k\) of the approximating spline of the form \(P(x)=\sum_{k=0}^NP_kp_k(x)+Q_kq_k(x)\) are now determined by boundary conditions and by collocation at the points \(\tau_j\), \(j=1,2,\cdots,2N\). This gives the system \({\mathbf Ac=f}\) of \(2N\) linear algebraic equations. For example, in the case of Dirichlet problem \({\mathbf A}=(a_{i,j})\), \(a_{i,j}=L[r_j](\tau_i)\), \(i,j=1,2,\cdots,2N\) \({\mathbf r}=[r_1,r_2, \cdots,r_{2N}]=[q_0,p_1,q_1,\cdots,p_{N-1},q_{N-1},q_N]\) and \({\mathbf c}=[Q_0,P_1,Q_1,\cdots,P_{N-1},Q_{N-1},Q_N]^T\). In each case of Dirichlet or Neumann conditions \({\mathbf A}\) is a band matrix of bandwidth 2. The authors observe that if we eliminate the variables \(P_k\) from each \(4\times 6\) block of equations corresponding to the block-diagonal of \({\mathbf A}\), and if we express them with the help of the variables \(Q_k\), then we get two systems of N equations. The first of them, for the variables \(P_k\), is tridiagonal, while the second one expresses the variables \(Q_k\) explicitely by the corresponding variables \(P_k\). In such a way we get a new, largely simplified algorithm. The paper contains numerical examples.