ROBP 三种本质相同的 PRG

这篇博文记录了 read-once branching program 三种本质相同的 pseudorandom generator —— [Nisan’92]、[Impagliazzo-Nisan-Wigderson’94] 和 [Nisan-Zuckerman’96]。

分别阐述了三种 PRG 的构造方法,最后用 Leftover Hash Lemma 这一桥梁将三者联系在一起。

引入

Read-once Branching Program

Nisan’s PRG

Pairwise-independent Hash Family 与其性质

构造

推论

Impagliazzo-Nisan-Wigderson PRG

Combinatorial Rectangle 与其 PRG

构造

更多 Open Problems

Nisan-Zuckerman PRG

Extractor 与其性质

构造

大一统 —— Leftover Hash Lemma

定理内容

Seedless Extractor 定义的是,要构造一个函数 Ext:{0,1}n{0,1}m\text{Ext} : \{0, 1\}^n \to \{0, 1\}^m,使得对于所有的 kk-source 都有 Δ(Ext(X),Um)<ε\Delta(\text{Ext}(X), U_m) < \varepsilon,这被证明是不可能的。

但如果我们换一个顺序,允许对于每种 kk-source 都换用不同的函数,那么情况会有所不同。

Thm

我们还可以将定理的结论换一个说法:Δ((H,H(X)),(UH,U[M]))<ε\Delta((H, H(X)), (U_H, U_{[M]})) < \varepsilon

Thm (Leftover Hash Lemma)H={h:{0,1}n{0,1}m}\mathcal{H} = \{h : \{0, 1\}^n \to \{0, 1\}^m \} 是一个 pairwise-independent hash family,并且 m=k2log(1/ε)m = k - 2 \log (1 / \varepsilon)。那么 Ext(x,h)h(x)\text{Ext}(x, h) \triangleq h(x) 是一个 strong (k,ε)(k, \varepsilon)-extractor。

Proof. 由 Cauchy 不等式可得:

Δ((H,H(X)),(Ud,Um))=12(H,H(X))(Ud,Um)1DM2(H,H(X))(Ud,Um)2=DM2(CP((H,H(X)))1DM)1/2\Delta((H, H(X)), (U_d, U_m)) = \dfrac{1}{2} \| (H, H(X)) - (U_d, U_m) \|_1 \\ \le \dfrac{\sqrt{D M}}{2} \| (H, H(X)) - (U_d, U_m) \|_2 = \dfrac{\sqrt{D M}}{2} \left( \text{CP}((H, H(X))) - \dfrac{1}{D M} \right)^{1 / 2}

其中 CP(X)Pr[X=X]=xPr[X=x]2\text{CP}(X) \triangleq \Pr[X = X'] = \sum_x \Pr[X = x]^2(这里 X,XX, X' i.i.d.)表示分布 XX 的碰撞概率 (collision probability)。

所以我们转而去控制 CP((H,H(X)))\text{CP}((H, H(X))),有:

CP((H,H(X)))=Pr[(H,H(X))=(H,H(X))]=Pr[H=H(X=X(XXH(X)=H(X)))]=Pr[H=H](Pr[X=X]+Pr[XX,H(X)=H(X)]) (独立性)=CP(H)(CP(X)+Pr[XX]Pr[H(X)=H(X)XX])CP(H)(CP(X)+Pr[H(X)=H(X)XX])1D(1K+1M)1+ε2DM\begin{aligned} \text{CP}((H, H(X))) & = \Pr[(H, H(X)) = (H', H'(X'))] \\ & = \Pr[H = H' \wedge (X = X' \vee (X \ne X' \wedge H(X) = H(X')))] \\ & = \Pr[H = H'] ( \Pr[X = X'] + \Pr[X \ne X', H(X) = H(X')] )\ (\text{独立性}) \\ & = \text{CP}(H) ( \text{CP}(X) + \Pr[X \ne X'] \cdot \Pr[H(X) = H(X') \mid X \ne X'] ) \\ & \le \text{CP}(H) ( \text{CP}(X) + \Pr[H(X) = H(X') \mid X \ne X'] ) \\ & \le \dfrac{1}{D} \left( \dfrac{1}{K} + \dfrac{1}{M} \right) \le \dfrac{1 + \varepsilon^2}{D M} \end{aligned}

综上:

Δ((H,H(X)),(Ud,Um))DM21+ε2DM1DM=ε2\Delta((H, H(X)), (U_d, U_m)) \le \dfrac{\sqrt{D M}}{2} \cdot \sqrt{\dfrac{1 + \varepsilon^2}{D M} - \dfrac{1}{D M}} = \dfrac{\varepsilon}{2}

即得。\Box

本质相同的论述

更多的讨论

参考文献

[HH23] Pooya Hatami and William M. Hoza. “Theory of Unconditional Pseudorandom Generators”. In: ECCC preprint TR23-019. 2023.

[INW94] Russell Impagliazzo, Noam Nisan, and Avi Wigderson. “Pseudorandomness for Network Algorithms”. In: Proc. 26th Annual ACM Symposium on Theory of Computing (STOC). 1994, pp. 356–364. isbn: 0897916638. doi: 10.1145/195058.195190

[Nis92] Noam Nisan. “Pseudorandom generators for space-bounded computation”. In: Combinatorica 12.4 (1992), pp. 449–461. doi: 10.1007/BF01305237

[NZ96] Noam Nisan and David Zuckerman. “Randomness is linear in space”. In: J. Comput. System Sci. 52.1 (1996), pp. 43–52. issn: 0022-0000. doi: 10.1006/jcss.1996.0004

[Vad12] Salil P. Vadhan. “Pseudorandomness”. In: Foundations and Trends in Theoretical Computer Science 7.1-3 (2012), pp. 1–336. doi: 10.1561/0400000010