Why does numerical matrix rank need a tolerance rather than exact comparison with zero?
MathematicsLevel: IntermediateRank and spectral numerics
Why does numerical matrix rank need a tolerance rather than exact comparison with zero?
Hints
- Exact algebra differs from floating point.
- Inspect singular-value scale.
Answer
Floating-point error creates tiny nonzero singular values. Tolerance separates numerical noise from resolved directions, depending on scale, precision and measurement uncertainty.
Source and licence: NumPy — numerical linear algebra · BSD-3-Clause · answer: Quantoros editorial