Superconvergence in collocation methods on quasi-graded meshes for functional differential equations with vanishing delays (Q2502317)

The paper begins with a brief review of earlier work by \textit{A. Bellen} [IMA J. Numer. Anal. 22, No.~4, 529--536 (2002; Zbl 1031.65089)] on the question of optimal local superconvergence in continuous implicit Runge-Kutta methods for the pantograph equation. Bellen's approach is extended to more general functional differential equations with nonlinear delay functions satisfying stated conditions on \(\theta\). The Volterra delay integro-differential equation (VDIDE) \[ y'(t)=f(t, y(t), y(\theta(t)), (Vy)(t), (V_\theta y)(t)), \] with initial condition \(y(0)=y_0\) is used to describe the ideas underlying the analysis. The authors state that the extension to general functional differential equations is obvious. The quasi-graded meshes to be used are introduced, along with the collocation method. Local Lagrange representations are used to write the collocation equation in a form suitable for the analysis and numerical computations to be performed. Two cases, \(\theta'(0)<1\) and \(\theta'(0)=1\), are considered separately. Sharp upper bounds are established for the attainable order of superconvergence. A theorem relating to the additional assumption needed when \(\theta'(0)=1\) is presented and proved. An example of a delay differential equation with a vanishing delay, \(\theta(t)=t-t^r (r \geq 2)\), is discussed. The authors comment that the number of necessary macro-intervals will become very large as \(r\) increases for such vanishing delays with a high order of tangency at \(t=2\). The open problem of determining the optimal superconvergence for second-order VDIDEs with vanishing delay (satisfying the same conditions) is briefly discussed. The paper concludes with a comment on the numerical implementation with regard to the tangency order \(r\). The authors indicate that results of further investiagation will be reported in a forthcoming pape\(r\).