Manuscript Oai265-B@2 · Oai265-B-v2.tex · section · 94b6f42f

Certified drift

Loss, heralding, drift, and reference safeguards › Certified drift · lines 958–988

Oai265-B@2#eq:driftbias

For independent nonidentically distributed scores, Hoeffding controls the average expectation, not an unidentified stationary channel.

If run $i$ in setting $j$ has

d_{⋄} (E_{j i}, E_{0}) \leq ξ_{j i},

the correlation changes by at most $2 ξ_{j i}$ . A valid additional fidelity allowance is

b_{drift} = \frac{1}{2} j = x, y, z \sum \frac{1}{n _{j}} i \sum ξ_{j i} .

Use $b_{cal} + b_{drift}$ . The iid binomial model must not automatically be applied to this drifting model.

An independently justified bound

d_{⋄} (E_{t}, E_{t_{0}}) \leq L ∣ t - t_{0} ∣

extends a point certificate by

e (E_{t}) \leq min {1, U_{t_{0}} + L ∣ t - t_{0} ∣} .

The constant $L$ is a separate input, not derived from the three correlations CheungHawk04B (:interval).

Displayed formulas in this passage

Oai265-B@2#eq:driftbias · line 969
$b_{drift} = \frac{1}{2} j = x, y, z \sum \frac{1}{n _{j}} i \sum ξ_{j i} .$
unlabeled · line 964
$d_{⋄} (E_{j i}, E_{0}) \leq ξ_{j i},$
unlabeled · line 979
$d_{⋄} (E_{t}, E_{t_{0}}) \leq L ∣ t - t_{0} ∣$
unlabeled · line 983
$e (E_{t}) \leq min {1, U_{t_{0}} + L ∣ t - t_{0} ∣} .$