A note on the efficient implementation of implicit methods for ODEs
arithmetic complexityefficient implementationimplicit Runge-Kutta methodsinner-outer iterationNewton iteration
Nonlinear ordinary differential equations and systems (34A34) Numerical methods for initial value problems involving ordinary differential equations (65L05) Multistep, Runge-Kutta and extrapolation methods for ordinary differential equations (65L06) Complexity and performance of numerical algorithms (65Y20)
The use of implicit methods such as implicit Runge-Kutta methods for the solution of ordinary differential equations (ODEs) requires the solution of systems of nonlinear algebraic equations of dimension \(sm\), where \(m\) is the size of the differential problem \[ y'= f(t,y),\quad y(t_0)= y_0\in\mathbb{R}^m.\tag{1} \] The system of algebraic equations has the form \[ F(y):= y- h(C\otimes I_m)f- g(y_0)= 0,\tag{2} \] where \(h\) is the stepsize, \(C\) is a full \((s,s)\)-matrix, \(I_m\) is the identity matrix of size \(m\), \(g:\mathbb{R}^m\to \mathbb{R}^{sm}\) is a known function and \(y= (y_1,\dots, y_s)^T\), \(f= (f_1,\dots, f_s)^T\), \(f_i= f(t_i, y_i)\) are the vectors with the unknown approximations to the solution at the grid points \(t_1,\dots, t_s\) when an implicit Runge-Kutta method with \(s\) stages is used. Usually the problem (2) is solved by using the modified Newton iteration \[ (I- hC\otimes J_0)\Delta y_i= F(y_{i-1}),\quad i= 1,2,\dots,\tag{3} \] where \(J_0\) is the Jacobian of the function \(f\) evaluated at \((t_0, y_0)\). The authors discuss a different approach using an inner iterative procedure for each outer iteration in (3). One such inner-outer iteration scheme is of the form \[ (I- hL\otimes J_0)\Delta y^{(j)}_i= (h(C- L)\otimes J_0)\Delta y_i^{(j- 1)}- F(y_{i- 1}),\quad j= 1,\dots,\mu,\tag{4} \] here \(\mu\) is a suitable parameter and the matrix \(L\) is obtained by the \(LU\) factorization of the matrix \(C\), where \(U\) has unit diagonal entries. The authors show that if all the entires \(l_i\) (the diagonal entries of the matrix \(L\)) are equal, the leading term in the arithmetic complexity is reduced from \({2\over 3}\cdot s\cdot m^3\) to \({2\over 3}\cdot m^3\).
- A special family of Runge-Kutta methods for solving stiff differential equations
- On the implementation of implicit Runge-Kutta methods
- On the potentiality of sequential and parallel codes based on extended trapezoidal rules (ETRs)
- Parallel linear system solvers for Runge-Kutta methods
- Runge-Kutta methods with a multiple real eigenvalue only
- Solving ordinary differential equations. II: Stiff and differential-algebraic problems.
- Triangularly Implicit Iteration Methods for ODE-IVP Solvers
- Recent advances in linear analysis of convergence for splittings for solving ODE problems
- Solving ordinary differential equations by generalized Adams methods: Properties and implementation techniques
- A simple ODE solver based on 2-stage Radau IIA
- Blended implementation of block implicit methods for ODEs
- Theoretically optimal inexact spectral deferred correction methods
- A note on efficient preconditioner of implicit Runge-Kutta methods with application to fractional diffusion equations
- Preconditioned iterative method for boundary value method discretizations of a parabolic optimal control problem
- Efficient implementation of Gauss collocation and Hamiltonian boundary value methods
- Efficient implementation of Radau collocation methods
- Analysis of a Chebyshev-based backward differentiation formulae and relation with Runge-Kutta collocation methods
- Blended implicit methods for the numerical solution of DAE problems
- Blended implicit methods for solving ODE and DAE problems, and their extension for second-order problems
- Acceleration of implicit schemes for large systems of nonlinear ODEs
This page was built for publication: A note on the efficient implementation of implicit methods for ODEs
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1379016)
