Superquadratic sensitivity separation: recursive joint zero #
Bounds the joint sensitivity of two indices q < q' of the recursive family at an input where both
are false. A bit changing both outputs makes some row clause true at the stronger index q',
hence also at the weaker index q, so it repairs a failed condition with respect to the candidate
lists of q: either of a target candidate (at most 15) or of a gate candidate (at most 3),
in which case the flip changes both gate predicates F (gateIndex q') and F (gateIndex q) on a
single child.
Main results #
recursive_joint_zero_at: at an input where both indices givefalse, the joint sensitivity is at most15 * sideSensitivity F false + 3 * jointSensitivity F true.recursive_joint_zero: the same bound for thefalse-side joint sensitivity of the recursive family.
References #
For a good labeling A, a nested family F, 0 < r and indices q < q' at which
recursiveFamily T A.label F is false on x, the joint sensitivity of the two predicates at x
is at most 15 * sideSensitivity F false + 3 * jointSensitivity F true.
Joint zero-side recurrence: for a good labeling, a nested family F and 0 < r,
jointSensitivity (recursiveFamily T A.label F) false is at most
15 * sideSensitivity F false + 3 * jointSensitivity F true.