Numerical Differentiation

Updated 15 Sep 2026

•
What it is
- •
  approximating a function's derivative using discrete function evaluations rather than symbolic differentiation
•
Forward difference
- •
  $f'(x) \approx \dfrac{f(x+h)-f(x)}{h}$ ; truncation error is order $O(h)$
- •
  total error bound (truncation + rounding): $E_{total}(h) \leq \dfrac{Mh}{2} + \dfrac{2\epsilon}{h}$ , where $M$ bounds $|f''(x)|$ and $\epsilon$ bounds the function's rounding error
•
Central difference
- •
  $f'(x) \approx \dfrac{f(x+h)-f(x-h)}{2h}$ ; truncation error is order $O(h^2)$ — a full order better than forward difference, since the odd-order Taylor terms cancel when subtracting $f(x-h)$ from $f(x+h)$
- •
  total error: $E_{central}(h) \approx \dfrac{M_3 h^2}{6} + \dfrac{\epsilon}{h}$ , where $M_3$ bounds $|f'''(x)|$
•
Second derivative (central difference)
- •
  $f''(t) \approx \dfrac{f(t+h)-2f(t)+f(t-h)}{h^2}$ ; dividing by $h^2$ amplifies input noise/uncertainty significantly (e.g. by 100× for $h=0.1$ ) — a caution when differentiating noisy sensor data
•
Related
- •
  Optimal Step Size
- •
  Richardson Extrapolation

Linked references 4