Home Wikituition Browse all terms Categories
Random term
Mathematical & Computational Physics

Round-off Error

The difference between the exact mathematical value of a function and the value computed using finite-precision floating-point arithmetic.

Governing formula Machine Epsilon ε_mach ≈ 2⁻⁵³ ≈ 1.11 × 10⁻¹⁶ (IEEE 754 Double Precision)
SI unit Floating-Point Precision Error

In depth

Computers store real numbers using finite bit lengths (e.g. 64-bit IEEE 754 double precision). Repeated arithmetic operations can accumulate round-off errors, causing catastrophic cancellation in subtraction of nearly equal numbers.

Examples in the real world

Minimizing catastrophic cancellation when computing quadratic formula roots for near-zero discriminant values.