The close relationships between methods for solving two-point boundary value problems. (English) Zbl 0575.65080
The authors describe close relationships between many methods for solving two-point boundary value problems. These relationships are useful for developing new implementations of methods like multiple shooting, stabilized march and invariant embedding. In § 2 of this paper the concept of conditioning of linear vector boundary value problems is given. It turns out to be dependent upon bounds for two quantities, one involving the boundary conditions and one the Green’s function of the problem. These results seem to be not quite new and may be dated back to work of M. Urabe [Funkc. Ekvacioj, Ser. Int. 9, 43-60 (1966; Zbl 0168.065)]. In § 3 it is shown why difficulties arise using superposition or reduced superposition for the numerical solution of linear boundary value problems. § 4 addresses multiple shooting and variants. The main result is that the inverse of the matrix M of the multiple shooting system is closely related to the Green’s function G(t,u) of the boundary value problem. In fact it mainly contains the values \(G(t_ i,t_ j)\) where the \(t_ i\) are the points partitioning the interval of definition. This result also seems to be not new but belonging to the folklore of numerical analysis, see for example the reviewer’s paper in Z. Angew. Math. Mech. 57, T 310-T 312 (1977; Zbl 0355.65058). Furthermore it is shown that certain undesirable properties of superposition are preserved when compactification is used to solve the multiple shooting system of equations. Other factorizations similar to invariant embedding avoid this particular difficulty. § 5 presents numerical examples.
Reviewer: H.Weber
MSC:
65L10 | Numerical solution of boundary value problems involving ordinary differential equations |
65L20 | Stability and convergence of numerical methods for ordinary differential equations |
34B15 | Nonlinear boundary value problems for ordinary differential equations |