Chopping vs Rounding

Updated 15 Sep 2026

•
What it is
- •
  two methods for representing a number with fewer significant digits: chopping (truncating) simply drops extra digits; rounding rounds to the nearest representable value
•
Accuracy comparison
- •
  rounding's maximum error bound is half that of chopping ( $\frac{1}{2} \times 10^{1-n}$ vs $10^{1-n}$ )
•
Worked example
- •
  $x = 2/3 = 0.666667...$ : chopped to 3 digits gives $0.666$ (error $0.100\%$ ); rounded gives $0.667$ (error $0.050\%$ )
•
Cumulative rounding error example
- •
  repeatedly summing $0.1$ (which cannot be represented exactly in binary — see Patriot Missile Failure) 100,000 times, with each term carrying a rounding error of $+9.5367 \times 10^{-8}$ , accumulates a total error of about $0.009537$ — the sum reads $10000.009537$ instead of the true $10000$
- •
  lesson: repeated summation of small binary-imprecise fractions accumulates bias linearly with the number of iterations, $O(N)$
•
Related
- •
  Error Taxonomy (Numerical Methods)
- •
  Machine Epsilon

Linked references 3