0
$\begingroup$

I was just doing some practice questions for a test, but have been stumped by the following for the past couple of hours.

I'm given a system such that: $$\frac{du}{dt} = v ~ ~ \& ~ ~ \frac{dv}{dt} = -f(u)$$

with Hamiltonian $$H = \frac{1}{2} \left(\frac{du}{dt}\right)^2 + \int f du.$$

I have to show that using the forward Euler method leads to a global error for $H$ that grows like $nh^2$ for step size $h$ and number of steps $n$.

I know that the global error can be calculated via $ \epsilon = |U^n - U(T)|$ but I'm not sure how to apply it in this case.

Thanks for any help!

EDIT: So if I understand correct, I have to calculate $|H_{n+1}-H_n| $

$\endgroup$
0
$\begingroup$

If you take the example $f(u)=u$, then you can interpret the Hamilton system in the complex plane as $$\dot z=\dot u+i\dot v = -i(u+iv)=-iz.$$

The Euler forward iteration thus produces elements $$ z_{k+1}=z_k+h(-iz_k)=(1-ih)z_k\implies z_n=(1-ih)^nz_0. $$

The Hamilton function is $H=\frac12(v^2+u^2)=\frac12|z|^2$, which gives on the Euler solution $$ H_n=\frac12|z_n|^2=\frac12(1+h^2)^n|z_0|^2=e^{nh^2+O(nh^4)}H_0 $$ so that indeed $H_n-H_0=nh^2e^{\frac12nh^2+O(nh^4)}H_0$.


Under more general conditions one gets \begin{align} H_{k+1}&=\frac12(v_k-hf(u_k))^2+F(u_k+hv_k) \\ &=H_k-hv_kf(u_k)+\frac12h^2f(u_k)^2 ~ + ~ f(u_k)(hv_k)+\frac12f'(u_k)(hv_k)^2+O(h^3) \\ &=H_k+\frac12h^2\bigl[f(u_k)^2+f'(u_k)v_k^2\bigr]+O(h^3) \end{align} Now you have to argue that the coefficient of the $h^2$ term remains bounded and the claim of the task follows.

$\endgroup$

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.