Richardson Extrapolation

Numerical Methods

Source inspiration: (Mathew 2000-2019).

Description

Richardson extrapolation combines two finite-difference estimates with different step sizes so that the leading truncation term cancels. In the legacy module, this is presented as a refinement of the central three-point derivative for \(f(x)=e^{-x}\sin(x)\) over \([0,\pi]\).

Using

\[ D_3(x,h)=\frac{f(x+h)-f(x-h)}{2h}, \]

the Richardson-improved estimate is

\[ D_5(x,h)=\frac{4D_3(x,h/2)-D_3(x,h)}{3}, \]

which upgrades the local truncation behavior from \(O(h^2)\) to \(O(h^4)\) for smooth functions.

Animations

Each animation below compares the base central difference with Richardson extrapolation for the legacy test function and interval.

Case 1 - profile comparison, \(D_3(x,h)\) vs. \(D_5(x,h)\)

Behavior: Both approximations converge as \(h\) shrinks, but the Richardson curve tracks \(f'(x)\) much more closely because the dominant \(h^2\) error term is canceled.

Julia source

Animation comparing D3 and Richardson D5 derivative curves against the true derivative over 0 to pi for decreasing h.

Case 2 - pointwise convergence at \(x_0=1\) on log-log axes

Behavior: The error slope for the three-point formula follows approximately \(O(h^2)\), while Richardson extrapolation follows approximately \(O(h^4)\) before roundoff effects appear.

Julia source

Animation of log-log error curves versus h at x0 equals 1 showing second-order and fourth-order convergence trends.

Derivation Notes

If \(D_3(x,h)=f'(x)+C_2 h^2 + C_4 h^4 + \cdots\), then

\[ \frac{4D_3(x,h/2)-D_3(x,h)}{3} = f'(x) + O(h^4), \]

because the \(C_2 h^2\) contribution cancels exactly.

References

Mathew, John H. 2000-2019. Numerical Analysis - Numerical Methods Modules. https://web.archive.org/web/20190808102217/http://mathfaculty.fullerton.edu/mathews/n2003/NumericalUndergradMod.html.