I understand almost all of this proof. But the part I can't find an explanation for is how they get the result for $r$ with the sums over the $\mathcal{O}(h^{\nu})$ all the way at the end of the proof. Can someone please explain this?

Page 1

Page 2

My problem is the following:

They define the global error as $\triangle z_{n+1} = \rho \triangle z_n + \delta_{n+1}$ with $\delta_{n+1} = \mathcal{O}{(h^{\min\lbrace p , q+1 \rbrace})}$ and $\rho = 1 - b^{T}A^{-1}\mathbb{1}$ (where $\rho$ is the stability function of the runge kutta method) then after repeated use of this formula they get:

$$\triangle z_{n} = \sum \limits_{i=1}^{n} \rho^{n-i} \delta_i$$

and from this alone they follow

$1) \triangle z_{n} = \mathcal{O}{(h^{\min\lbrace p , q+1 \rbrace})} \mbox{ for } -1 \leq \rho < 1$

$2) \triangle z_{n} = \mathcal{O}{(h^{\min\lbrace p-1 , q \rbrace})} \mbox{ for } \rho = 1$

$3) \mbox{The solution diverges} \mbox{ for } \rho > 1$

And I don't understand how they get these answers.


Note that $$|\Delta z_n| \le C h^{\min\{p, q+1\}} \left|\sum_{i=1}^n \rho^{n-i}\right|.$$

  • If $|\rho| < 1$, then the sum can be bounded by the geometric series $1 + \rho + \rho^2 + \cdots = \frac{1}{1-\rho}$.
  • If $\rho = -1$, then the sum can be bounded by $1$.
  • If $\rho = 1$, then the sum equals $n \le C/h$ (the relationship between $n$ and $h$ is an assumption stated in the theorem).
  • $\begingroup$ Thanks for your answer, your explanation made it much clearer $\endgroup$ – mmg97 Jan 20 at 19:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.