3
$\begingroup$

In Lindell's tutorial "How to simulate it" [2016/046], section 4.3, he gives a semi-honest protocol for oblivious transfer, based on enhanced trapdoor permutations and a corresponding hard-core predicate. I don't understand some details in the proof and hope somebody can explain it to me.

He gives the following definition: $B$ is a hard-core predicate of a trapdoor permutation $f_\alpha$ if:

for every non-uniform PPT adversary $A$ there exists a negligible function $\mu(n)$ such that for all $n$: $$ \Pr[A(1^n, \alpha, r) = B(\alpha, f_\alpha^{-1}(S(\alpha; r)))] \leq \frac{1}{2} + \mu(n), $$

where $\alpha$ is an index in the collection of functions $\{f_ \alpha\}$ and $S$ samples (almost uniformly) in the range/domain of $f_\alpha$.

Protocol 4.2 computes OT-functionality $((b_0,b_1), \sigma) \mapsto (\lambda, b_\sigma)$ as follows:

  • $P_1$ sends index $\alpha$ of a trapdoor function $f_\alpha$, where $P_1$ knows the trapdoor but $P_2$ does not.
  • $P_2$ samples random $x_\sigma$ and $y_{1-\sigma}$, computes $y_\sigma=f_\alpha(x_\sigma)$, and sends $(y_0, y_1)$.
  • $P_1$ uses the trapdoor to compute $x_0$ and $x_1$, then sends $B(\alpha, x_0) \oplus b_0$ and $B(\alpha, x_1) \oplus b_1$.
  • $P_2$ computes $B(\alpha, x_\sigma)$ and can therefore output $b_\sigma$.

Part of the proof of security (Theorem 4.3) constructs a simulator $S_2$ for corrupt party $P_2$ (the recipient), by showing that if you can distinguish the output of $S_2$ from that of $P_2$, you can use that to construct an algorithm $A$ that breaks the assumption that $B$ is a hard-core predicate, summarized as follows:

Algorithm $A$ [...] receives $(1^n, \alpha, r)$ for input. $A$'s aim is to guess $B(\alpha, S(\alpha; r))$.

However, there is no guessing because you can simply compute $B(\alpha, S(\alpha; r))$, right?

The final equation of the proof computes $\Pr[A(1^n, \alpha, r) = B(\alpha, x)]$, but it is unclear to me here what $x$ is in this expression (the preceding paragraph only mentions how to sample $x_\sigma$). Given the definition of the hard-core predicate, I would have expected to see $f_\alpha^{-1}$ in this part of the proof.

$\endgroup$
7
$\begingroup$

The answer is that it's me and not you. There was a typo in the proof and indeed $x$ referred to here is $f^{-1}_\alpha(S(\alpha,r))$. I have made the change in the proof and updated the ePrint paper. Thank you for pointing out the typo.

$\endgroup$
2
  • $\begingroup$ Great! Thank you for writing the tutorial. I'm still going through it but so far it has been very helpful. $\endgroup$
    – Sebastian
    Apr 19 at 14:38
  • 1
    $\begingroup$ You are most welcome! I am really glad that it's helpful. $\endgroup$ Apr 19 at 14:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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