Tribes under bounded independence #
The Tribes function Tribes_{w,s} (Cslib.BitString.tribes) is computed by a depth-two circuit of
the unbounded fan-in basis, an OR of s ANDs of w inputs each (tribesCircuit), with
s + 1 gates and fan-in max s w. Through
Chattopadhyay–Zuckerman, Theorem 4.1
(Cslib.Circuits.Boolean.Braverman.isKWiseResilient_of_isResilient), the resilience of Tribes
(Cslib.BitString.isResilient_tribes) and its near balance
(Cslib.BitString.abs_expect_bitVal_tribes_sub_half_le) carry over, with an error 2 ^ (-p), to
every t-wise independent distribution with t ≥ (64 p) ^ 157, where p ≥ 2, 2 s + 8 ≤ 2 ^ p
and w < 2 ^ p; for the bias, t ≥ (64 p) ^ 64 suffices, where p ≥ 1 and s, w < 2 ^ p. These
exponents are those of the library's form of Braverman's theorem at depths 5 and 2; they are
ours, and Chattopadhyay and Zuckerman's exponent 3 d + 6 is not claimed.
Main definitions #
Cslib.Circuits.Boolean.tribesCircuit: the depth-two circuit computing Tribes.
Main results #
Cslib.Circuits.Boolean.eval_tribesCircuit,Cslib.Circuits.Boolean.size_tribesCircuit,Cslib.Circuits.Boolean.depth_tribesCircuit_le,Cslib.Circuits.Boolean.fanInAtMost_tribesCircuit: its function, size, depth and fan-in.Cslib.Circuits.Boolean.Braverman.isKWiseResilient_tribes: Tribes is resilient under bounded independence.Cslib.Circuits.Boolean.Braverman.abs_sum_bitVal_tribes_sub_half_le: Tribes is nearly balanced under bounded independence.
References #
The depth-two circuit computing Tribes_{w,s}: the OR of the ANDs of the tribes.
Instances For
tribesCircuit w s computes Tribes_{w,s}.
tribesCircuit w s has s + 1 gates.
tribesCircuit w s has depth at most two.
tribesCircuit w s has fan-in at most max s w.
Tribes is resilient under bounded independence, our instance of
Chattopadhyay–Zuckerman, Theorem 4.1: if p ≥ 2,
2 s + 8 ≤ 2 ^ p and w < 2 ^ p, then Tribes_{w,s} is t-independent
(q, 2 ^ (q - w) + 2 ^ (-p))-resilient for every t ≥ (64 p) ^ 157. The exponent 157 is
(2 d + 9) (d + 10) + 1 at the depth d = 2 of tribesCircuit.
Tribes is nearly balanced under bounded independence, our instance of
Chattopadhyay–Zuckerman, Theorem 4.1, bias part: if s is largest
with (1 - 2 ^ (-w)) ^ s ≥ 1 / 2 O'Donnell, Definition 4.11, p ≥ 1 and
s, w < 2 ^ p, then under every t-wise independent distribution with t ≥ (64 p) ^ 64,
Tribes_{w,s} is true with probability within 2 ^ (-w) + 2 ^ (-p) of 1 / 2. The exponent 64
is (2 d + 3) (d + 7) + 1 at d = 2.