I'd like to prove that the order of the two-steps Backward Differentiation Formula is $O(k^2)$. For this, I write $$y'_i\approx \dfrac{y_{i}-{\tfrac {4}{3}}y_{i-1}+{\tfrac {1}{3}}y_{i-2}}{2k/3}=\dfrac{1}{k}\left(\dfrac{3}{2}-2E^{-1}+\dfrac{1}{2}E^{-2}\right)=\dfrac{1}{k}\left(\nabla+\dfrac{\nabla^2}{2}\right).$$
So, the approximation is $$D= \dfrac{1}{k}\left(\nabla+\dfrac{\nabla^2}{2}+\dfrac{\nabla^3}{3}+\dfrac{\nabla^4}{4}+\cdots\right)\approx \dfrac{1}{k}\left(\nabla+\dfrac{\nabla^2}{2}\right)$$ and the error $$ \dfrac{1}{k}\left(\dfrac{\nabla^3}{3}+\dfrac{\nabla^4}{4}+\cdots\right)$$
How could I proceed?
Thank you in advance!
\Delta$\Delta$ symbol, not the gradient\nabla$\nabla$. /// You could be more explicit in writing that here $\Delta=1-E^{-1}$, and $E=e^{kD}$, so that $kD=-\ln(1-\Delta)=$ logarithm MacLarin-Taylor expansion. With all that, your approach is correct. You could, alternatively, also insert the exponential or Taylor shift series for $E^{-1}=e^{-kD}$ directly. – Lutz Lehmann Aug 16 '22 at 07:11