Skip to content

Fields, grades and numerical arithmetic

There are several places where arithmetic enters a persistence computation. The coefficient field belongs to the vector spaces and maps; parameter grades describe when objects enter; numerical tolerances decide how a floating-point calculation treats nearly dependent vectors. Consider a differential given by multiplication by two. Over the rational numbers it has zero kernel; over the field with two elements it is the zero map. Changing the field can change the answer, while changing a basis within one field does not.

The guide to comparing results puts coefficient choices beside changes of basis and input construction. It connects each comparison to the spaces and maps being compared, so a shared plot has a mathematical interpretation. Coefficient arithmetic belongs to those spaces and maps; geometric distances and parameter grades have their own arithmetic.

Numerical derived algebra examines what happens when rank and related calculations use numerical tolerances. A singular value near a threshold can change the reported rank and therefore a later algebraic answer. The guide helps you inspect that choice, distinguish a small residual from an exact identity and decide what evidence the computation provides.

For parameter coordinates, exact rhomboid grades supplies a complementary example. Two distinct events can have the same displayed floating-point coordinate. Keeping their grades exact preserves the membership query even though the picture rounds them to one location. This is why the word “exact” needs an object: exact coefficients, exact grades and an exact optimization answer describe different parts of a calculation.