The executable rational ceiling majorizes the complex norm of the rational coefficient.
The coordinate majorant bounds the complex norm at the selected field embedding.
Normalizing an evaluation eliminant preserves nonzeroness.
Normalizing an evaluation eliminant preserves every nonzero complex root.
The normalized evaluation eliminant gives the lower bound used by the bounded zero test.
One Horner step preserves the executable error-majorant recurrence.
A ball that passes the executable zero-exclusion test cannot contain zero.
A sufficiently small ball containing a value above a reciprocal lower bound passes the executable zero-exclusion test.
The certified fixed-field Horner evaluator has the radius promised by its executable error majorant.
The executable radius predicate is its stated real inequality.
The real radius inequality implies the executable radius predicate.
The prescribed endpoint has enough logarithmic slack: any ball whose
radius is bounded by max 1 majorant ulps at that precision satisfies the
zero-test radius predicate.
A successful bounded precision search exposes an available ball satisfying the requested radius predicate.
The bounded search succeeds whenever its prescribed endpoint produces a ball satisfying the radius predicate.
The prescribed bounded precision search succeeds for the fixed-field ball evaluator.
The bounded zero-retention test cannot fail for the fixed-field ball evaluator.
Conditional correctness of the bounded zero test. Whenever the search returns, it retains exactly the zero value represented by the eliminant root and the certified evaluation balls.