Cryptography Stack Exchange is a question and answer site for software developers, mathematicians and others interested in cryptography. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Reading Goldreich's Foundations of Cryptography II, I found this proof for the security of the common pseudorandom generator + XOR encryption scheme (Proposition 5.2.12 in the book):

Assume you have a secure pseudorandom generator $g \colon \{ 0,1\}^n \to \{0,1\}^m$, and you construct your encryption scheme in the standard way by XOR'ing the output of $g$ to a message of length $m$. To prove that this is one-message-IND-secure, Goldreich proceeds as follows:

  1. Assume, for the sake of contrdiction, that there is an adversary $\mathcal{A}$ (really a poly-sized circuit in the book, but that is not important here) against this encryption scheme, such that for $x,y \in \lbrace 0,1 \rbrace^m$, $U_n \in \lbrace 0,1 \rbrace^n$ and polynomial $p$ we have: $$ |Pr[\mathcal{A}(x \oplus g(U_n)) = 1] - Pr[\mathcal{A}(y \oplus g(U_n)) = 1] > \frac{1}{p(n)}.$$
  2. Observe that for $U_m \in \lbrace 0,1 \rbrace^m$ we have: $$ Pr[\mathcal{A}(x \oplus U_m) = 1] = Pr[\mathcal{A}(y \oplus U_m) = 1].$$
  3. Then, WLOG: $$ |Pr[\mathcal{A}(x \oplus g(U_n)) = 1] - Pr[\mathcal{A}(x \oplus U_m) = 1]| > \frac{1}{2p(n)},$$ which contradicts the security of $g$ (since we can distinguish its output from random).

My question

How do you get to that last inequality in step 3? In particular, how do you derive the term $\frac{1}{2p(n)}$?

share|improve this question
up vote 5 down vote accepted

The inequality is obtained by a distance argument.

Consider two points $X,Y$ on the real line. Taking another point $Z$, you have $|X-Z| + |Y-Z| \geq |X-Z+Z-Y| = |X-Y|$.

Applying this "triangle" inequality to your equality 1, we have for any $z \in \mathbb{R}$, $\begin{array}{l} \bigl\lvert\Pr[A(x\oplus g(U_n))=1] - z\bigr\rvert + \bigl\lvert \Pr[A(y\oplus g(U_n))=1]- z\bigl\rvert \\ \geq \bigl\lvert\Pr[A(x\oplus g(U_n))=1] - z + z - \Pr[A(y\oplus g(U_n))=1]\bigr\rvert \\ = \bigl\lvert\Pr[A(x\oplus g(U_n))=1]-\Pr[A(y\oplus g(U_n))=1]\bigr\rvert \\ > 1/p(n). \\ \end{array}$

By setting $z = \Pr[A(x \oplus U_m)=1] = \Pr[A(y \oplus U_m) = 1]$, we have $\begin{array}{l} \bigl\lvert\Pr[A(x\oplus g(U_n))=1] - \Pr[A(x \oplus U_m)=1]\bigr\rvert \\ + \bigl\lvert \Pr[A(y\oplus g(U_n))=1]-\Pr[A(y \oplus U_m) = 1]\bigr\rvert \\ > 1/p(n). \\ \end{array}$

The one of two absolute values is bigger than $1/(2p(n))$. Therefore, WLOG, you obtain your equation 3.

share|improve this answer
Exactly the answer I was looking for, thx! It's funny though: I had done exactly the same thing as you, including the consideration of the "triangle" inequality. But, when I got to your last equation, I couldn't figure out where that $1/2$ factor should come from. However, its good too see that I wasn't to far of :) – hakoja Mar 16 '13 at 15:35
Suppose the two absolute values are at most 1/(2p), then you get 1/p < |...| + |...| <= 2 * 1/(2p) = 1/p, that is, contradiction. – xagawa Mar 20 '13 at 16:12

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.