claimroutineinferred_dep
Noise part needs no sum over overflow groups
True-type overflow noise error needs no sum over overflow groups
true-type noise term of the error budget
Statement (paper)
Since on each row at most one first-overflow event occurs, pointwise, regardless of dependence between noise and overflow. Hence, by the absolute prefix-noise bound, the noise part of the true-type first-overflow error has norm at most , without an extra sum over .
In plain English
Because each row has at most one first-overflow event , the indicator-weighted noise sum is pointwise at most , whatever the dependence between noise and overflow. So the noise part of the true-type first-overflow error has norm at most , with no extra sum over .
Why it's here
Relative stability gives an error of the form once overflow first occurs in group . Since the events are disjoint on each row, summing over costs nothing extra. No independence between noise and overflow is needed, and the noise contribution stays a single geometric sum that the later convergence step can control.
Classified routine:
Follows from the disjointness of the first-overflow events on each row: the indicator-weighted sum is pointwise bounded by the full sum. Combined with the earlier relative-stability lemma and the absolute prefix-noise bound, this is simple bookkeeping.
Uses
Used by
Sources
PaperOpen
- Section
- §8.4 Charging overflow events
- Page
- —
- Label
—- Line
sections/compression.tex:518
Leannot aligned
Phase 2 aligns this node to a Lean declaration.
Friction
- infoinferred_depoutline: "Noise part via uniqueness of first overflow" rests on lem:buf-relative-stability