0
$\begingroup$

I have an exercise that I do not understand. We have to solve an initial value problem:

$$ \begin{array}{ccl} y'(t) &=& f(t,y(t)) \\ y(a) &=& y_0 \end{array} $$

We have to derive an effective method to calculate $\alpha$, $\beta$ and $\gamma$ with the highest possible accuracy. The method has the form below: $$ y_{k+1} = \alpha y_k + \beta y_{k-1} + \gamma h f(t_k, y_k) $$ We can even assume that $y_1 = y(a+h)$ and that the interval is equidistant.

I can guess that we have to derive a Euler forward, but I cannot solve it? Besides, they ask the order of accuracy of the found method. Shall I write all the $y(a+h)$ as a Taylor polynomial? I would greatly appreciate some guidance.

$\endgroup$
  • $\begingroup$ Writint out a Taylor polynomial would ideed be a good idea, I think. $\endgroup$ – Matti P. May 22 at 7:14
  • 1
    $\begingroup$ No, it will not be Euler forward, the resulting method is a second order two-step midpoint or Nyström method, like in this question, further explored here. $\endgroup$ – LutzL May 22 at 7:24
  • $\begingroup$ I don't think it's a Nyström or modified Euler, it's a beginner course and we did not do those methods. I tried to make a Taylor polynomial but nothing disappears. $\endgroup$ – Dovendyr May 22 at 9:25
  • $\begingroup$ Simple order 2 methods are absolutely suited for a beginner course. Besides, this is more an exercise in Taylor polynomials than in numerical integration. $\endgroup$ – LutzL May 22 at 9:31
  • $\begingroup$ I still don't understand, can you please show what you mean and which parts I am supposed to develop with Taylor? $\endgroup$ – Dovendyr May 26 at 7:24
0
$\begingroup$

For the test problem $f(t,y)=y$ you get the exact solution $y=Ce^x$ and with the constant chosen so that $y_k=1$ the equation reduces to $$ e^h=α+βe^{-h}+γh $$ which has to be correct up to some power of $h$.

Inserting the power series of the exponential and comparing coefficients of equal degree gives \begin{align} γh&=(1+h+\tfrac12h^2+\tfrac16h^3+...)-α-β(1-h+\tfrac12h^2-\tfrac16h^3\pm...) \\[1em]\hline 0&=1-α-β\\ γ&=1+β\\ 0&=\tfrac12(1-β)\\ 0&=\tfrac16(1+β) \end{align} which have to be satisfied from top down as far as possible. Obviously, the last equation contradicts the next-to-last one. Thus $β=1$, $γ=2$ and $α=0$.

You will need to confirm that this holds true for any differential equation, not just this (quite standard) test case.

$\endgroup$

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.