Stability ordinates of Adams predictor-corrector methods (Q747638)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Stability ordinates of Adams predictor-corrector methods |
scientific article |
Statements
Stability ordinates of Adams predictor-corrector methods (English)
0 references
19 October 2015
0 references
The article contains a qualitative analysis of the root locus curve (see [\textit{E. Hairer} and \textit{G. Wanner}, Solving ordinary differential equations. II: Stiff and differential-algebraic problems. Reprint of the 1996 2nd revised ed. Reprint of the 1996 Berlin: Springer (2010; Zbl 1192.65097), p. 242]) for a sub-set of multistep numerical methods which, when applied to the Dahlquist's test problem \[ \frac{dy}{dx}=\lambda y,\tag{1} \] can be expressed in the form of: \[ y_n=y_{n-1}+\lambda h\left(\sum^m_{k=0}\alpha_k\Delta^k y_{n-1}+\sum^m_{k=0}\beta_k \Delta^k y_n\right),\, \alpha_k,\beta_k\in\mathbb R.\tag{2} \] In particular, the authors consider Adams-Bashforth (AB), Adams-Moulton (AM) and Adams predictor-corrector methods which are combinations of the previous two: \(\mathrm{AB}(p-1)\) -- \(\mathrm{AM}p\), \(\mathrm{AB}p\) -- \(\mathrm{AM}p\). The main question addressed in the article is: For which orders \(p\) do the mentioned methods possess a property of ``nonzero imaginary stability boundary'' (nonzero ISB), which is, according to the definition in the article, equivalent to the fact that some part of the root locus curve of a method goes to the right of the imaginary axis? The theoretical framework used in the article consists of the following facts: {\parindent=0.6cm\begin{itemize}\item[1.] The root locus curve of a given method is described by the complex-valued function of a real variable \(\theta\): \[ \xi(\theta)=\frac{\rho(e^{i\theta})}{\sigma(e^{i\theta})},\, \theta\in [0, 2\pi],\tag{3} \] where \(\rho(r)\) and \(\sigma(r)\) -- the generating polynomials of the method (see, for example [Hairer and Wanner, loc. cit., p. 240]).\item[2.] As follows from the Theorem 2.1 from \textit{A. Iserles} [A first course in the numerical analysis of differential equations. 2nd ed. Cambridge: Cambridge University Press (2009; Zbl 1171.65060)] if the given multistep method is exact, then \[ \frac{\rho(e^{i\theta})}{\sigma(e^{i\theta})}=\ln(e^{i\theta})=i\theta;\tag{4} \] whereas if the numerical scheme is of order \(p\) then \[ \frac{\rho(e^{i\theta})}{\sigma(e^{i\theta})}=i\theta+c_p(i\theta)^{p+1}+d_p (i\theta)^{p+1}+\mathcal O((i\theta)^{p+3});\tag{5} \] \item[3.] for the given multistep method (2) its ``exact'' analogous can be expressed in the form: \[ y_n=y_{n-1}+\lambda h\left(\sum^{+\infty}_{k=0}\alpha_k\Delta^ky_{n-1}+\sum^{+\infty}_{k=0}\beta_k\Delta^ky_n\right).\tag{6} \] \end{itemize}} The theorems stated and proved in the article reveal the relationship between the order \(p\) of a method and the fact that the method has (or does not have) a nonzero ISB. The general approach used to prove statements of the theorems can be briefly described as the following: {\parindent=6mm \begin{itemize}\item[1.] Using \(\xi =\lambda h\), and combining (2) with (3) we get \[ e^{in\theta}=y_{n-1}+\xi\left(\sum^m_{k=0}\alpha_k\Delta^k y_{n-1}+\sum^m_{k=0}\beta_k \Delta^k y_n\right);\tag{7} \] \item[2.] next, combining (4) with (6) we get \[ e^{in\theta}=y_{n-1} + i\theta\left(\sum^{+\infty}_{k=0}\alpha_k\Delta^ky_{n-1}+\sum^{+\infty}_{k=0}\beta_k\Delta^ky_n\right);\tag{8} \] \item[3.] subtracting (8) from (7) and using expansion (5) we obtain \[ \begin{multlined} [c_m(i\theta)^{m+2} +d_m(i\theta)^{m+3}+\mathcal O(i\theta)^{m+4}]\left(\sum^m_{k=0}\alpha_k\Delta^k y_{n-1}+\sum^m_{k=0}\beta_k \Delta^k y_n\right)\\ =i\theta\left(\sum^{+\infty}_{k=m+1}\alpha_k\Delta^ky_{n-1}+\sum^{+\infty}_{k=m+1}\beta_k\Delta^ky_n\right).\end{multlined} \] Collecting like powers of \(i\theta\) and analysing representations of \(c_m\) and \(d_m\) we make a conclusion about for which \(m\) expression \[ c_m (i\theta)^{m+2}+d_m (i\theta)^{m+3} \] has positive real part that, in turn, corresponds to nonzero ISB of the method in question. \end{itemize}} In general all the statements presented in the article are followed by thorough proofs. Almost each theoretical result is supported by a comprehensive example. The last paragraph of the article is devoted to the applications of the obtained theoretical results to numerical analysis of the one-dimensional wave equation.
0 references
Adams methods
0 references
predictor-corrector
0 references
imaginary stability boundary
0 references
linear multistep methods
0 references
finite difference methods
0 references
stability region
0 references
stability ordinate
0 references
numerical example
0 references
root locus curve
0 references
Adams-Bashforth
0 references
Adams-Moulton
0 references
wave equation
0 references