Parallel iteration of the extended backward differentiation formulas
algorithmconvergenceextended backward differentiation formulasinitial value problemiteration methodnumerical experimentsparallel computationpredictor-corrector method
Nonlinear ordinary differential equations and systems (34A34) Numerical methods for initial value problems involving ordinary differential equations (65L05) Finite difference and finite volume methods for ordinary differential equations (65L12) Stability and convergence of numerical methods for ordinary differential equations (65L20) Error bounds for numerical methods for ordinary differential equations (65L70) Parallel numerical computation (65Y05)
The subject of this paper is the proposition of a parallelized algorithm realizing the so called extended backward differentiation formulas (EBDF) and the modified extended backward differentiation formulas (MEBDF). For the initial value problem for ordinary differential equations of the form NEWLINE\[NEWLINE{dy\over{dt}}=f(y)\;\;\;t\geq t_0, \tag{1}NEWLINE\]NEWLINE NEWLINE\[NEWLINEy(t_0)=y_0,NEWLINE\]NEWLINE the EBDF formulas by \textit{J. R. Cash} [Numer. Math. 34, 235-246 (1980; Zbl 0411.65040)] for evaluation of \(y_{n+1}\) are of the following form NEWLINE\[NEWLINEy_{n+1}=\sum_{j=1}^ka_jy_{n-j+1}+h[b_0f(y_{n+1})+b_1f(y_{n+2})]. \tag{2}NEWLINE\]NEWLINE In order to compute approximately the starting vector \(u_{n+1}\) for \(y_{n+1}\) and the vector \(y_{n+2}\), the standard implicit BDF corrector (playing here the role of a predictor) has to be used twice: NEWLINE\[NEWLINEu_{n+1}=\sum_{j=1}^k{\bar a}_jy_{n-j+1}+h{\bar b}_0f(u_{n+1}),NEWLINE\]NEWLINE NEWLINE\[NEWLINEu_{n+2}=\sum_{j=1}^k{\bar a}_jy_{n-j+2}+h{\bar b}_0f(u_{n+2}).NEWLINE\]NEWLINE The modified formulas differ a little. Since the method is implicit, the iteration is needed to compute vectors \(y_{n+1}\), \(u_{n+1}\) and \(u_{n+2}\). Parallelization enters at the level of resolution of the resulting system of the, in general, nonlinear equations. The iterative process is organized so that vectors \(u_{n+1}\), \(u_{n+2}\), \(y_{n+1}\) are computed simultaneously. The rate of convergence of the iterative process is discussed. The paper contains results of numerical experiments. Also timing of various versions of algorithms are compared.
- Diagonalizable extended backward differentiation formulas
- The computation of EBDF and MEBDF methods using backward differences
- Generalization of backward differentiation formulas for parallel computers
- An MEBDF code for stiff initial value problems
- Solving time dependent PDEs via an improved modified extended BDF scheme
- Generalized linear multistep methods for ordinary differential equations
- Passage to the differences for BDF methods
- Modified extended backward differentiation formulae for the numerical solution of stiff initial value problems in ODEs and DAEs
- The MOL solution of time dependent partial differential equations
- The integration of stiff systems of ODEs using NDFs and MEBDFs
- Diagonalizable extended backward differentiation formulas
- Generalization of backward differentiation formulas for parallel computers
- A novel class of collocation methods based on the weighted integral form of ODEs
- Adapted BDF algorithms: Higher-order methods and their stability
- Efficient time integrators in the numerical method of lines
- Multi-implicit peer two-step W-methods for parallel time integration
- Exponentially fitted two-step Runge-Kutta methods: construction and parameter selection
- scientific article; zbMATH DE number 221341 (Why is no real title available?)
- The computation of EBDF and MEBDF methods using backward differences
- Exponential fitting BDF-Runge-Kutta algorithms
- Exponential fitted Gauss, Radau and Lobatto methods of low order
This page was built for publication: Parallel iteration of the extended backward differentiation formulas
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2713133)
