Suppose that we play the game from Adversarial Indistinguishability but adversarial can choose three messages $m_0, m_1, m_2$. Of course, $Pr[M=m_i]=1/3$ for $i=0,1,2$. I suppose that to have adversarial indistinguishability, one cannot have an advantage greater than $1/3$. The question is if this is stronger than the version with two messages. Intuitively it is, but then we could take more and more messages and make the advantage of adversarial lesser and lesser. Is it neccessary? For some reason, in the definition, there are two messages - is this enough?
1 Answer
Remark: here I'm using the indexes $0,1,2$ and $0,1$ instead of $1, 2, 3$ and $1,2$.
We have to show the $3$-indistinguishability problem is equivalent to the $2$-indistinguishability one.
$2$-indistinguishability is easier than $3$-indistinguishability.
First let consider it exists an adversary $\mathcal{A}_3$ against the $3$-indistinguishability problem with advantage $\epsilon$.
Define $\mathcal{B}^{\mathcal{A}_3}_2$:
Receive three messages $(m_0, m_1, m_2)$ from $\mathcal{A}_3$
$x \xleftarrow{\\\$} \mathbb{Z}_3$
$(m^\prime_0, m^\prime_1) := (m_{(1+ x \mod 3)}, m_{(2+ x \mod 3)})$
Send $(m^\prime_0, m^\prime_1)$ to the challenger, and receive $c$.
Send $c$ to $\mathcal{A}_3$ and receive $b$.
If $b=x$ then return a random bit $b^\prime$ else return $(b- x \mod 3)$.
We first prove that $\mathcal{B}_2$ has probability to win $\frac{1-\epsilon}{4} +\epsilon$.
Let call $b''$ the bit chosen by the challenger.
$\mathbb{P}(\mathcal{B}_2 wins)= \mathbb{P}(b- x \mod 3 = b'')\mathbb{P}(\mathcal{B}_2 wins| b- x \mod 3 = b'') + \mathbb{P}(b- x \mod 3 \neq b'')\mathbb{P}(\mathcal{B}_2 wins| b- x \mod 3 \neq b'')$
$= \epsilon \cdot 1 + (1 - \epsilon)\mathbb{P}((b=x) \wedge b^\prime =b'' | b- x \mod 3 \neq b^{\prime\prime}) $
$= \epsilon + (1 - \epsilon)\frac{1}{2}\cdot\frac{1}{2}.qed$
Now we can look the advantage: $|\frac{1}{2} - (\frac{1-\epsilon}{4} +\epsilon)|=|\frac{1- 3 \epsilon}{4}| = \frac{1}{12}|\frac{1}{3}- \epsilon|$.
If $|\frac{1}{3}- \epsilon|$ is non negligible, it implies $|\frac{1}{2} - (\frac{1-\epsilon}{4} +\epsilon)|$ is also non negligible.
$3$-indistinguishability is easier than $2$-indistinguishability.
Now let consider it exists an adversary $\mathcal{A}_2$ against the $2$-indistinguishability problem with advantage $\epsilon$.
Define $\mathcal{B}^{\mathcal{A}_2}_3$:
Receive two messages $(m_0, m_1)$ from $\mathcal{A}_2$
$b \xleftarrow{\\\$} \mathbb{Z}_2$
$m_2 := m_{b}$
Send $(m_0, m_1, m_2)$ to the challenger, and receive $c$.
Send $c$ to $\mathcal{A}_2$ and receive $b^\prime$.
$x \xleftarrow{\\\$} \big\{b, 2\big\}$
If $b^\prime=b$ then return $x$ else return $b^\prime$
We first compute the probability to win for $\mathcal{B}_3$.
$\mathbb{P}(\mathcal{B}_3 wins) = \frac{1}{3}\mathbb{P}(\mathcal{B}_3 wins|b''=2) + \frac{1}{3}\mathbb{P}(\mathcal{B}_3 wins| b''=b)+ \frac{1}{3}\mathbb{P}(\mathcal{B}_3 wins| b''=1 - b) $
$\mathbb{P}(\mathcal{B}_3 wins) = \frac{1}{3}\mathbb{P}(b'=b \wedge x=2|b''=2) + \frac{1}{3}\mathbb{P}(b'=b \wedge x=b| b''=b)+ \frac{\epsilon}{3}.$
Because $x$ is picked independently of $b'$:
$\mathbb{P}(\mathcal{B}_3 wins)$ $= \frac{1}{3}\mathbb{P}(b'=b|b''=2) \cdot \mathbb{P}(x=2|b''=2) + \frac{1}{3}\mathbb{P}(b'=b|b''=b') \cdot \mathbb{P}(x=b''|b''=b')+ \frac{\epsilon}{3} $
$\mathbb{P}(\mathcal{B}_3 wins) = \frac{1}{3}\epsilon \cdot \frac1 2 + \frac{1}{3}\epsilon \cdot \frac1 2+ \frac{\epsilon}{3} = \frac{2\epsilon}{3}.$
Now we compute the advantage of $\mathcal{B}_3$: $|\frac1 3 - \frac{2\epsilon}{3} |= \frac1 6 |\frac 1 2 - \epsilon|.$
If $|\frac{1}{2}- \epsilon|$ is non negligible, it implies $|\frac{1}{3} - \frac{2\epsilon}{3}|$ is also non negligible.
We deduce these problems are equivalent.
-
1$\begingroup$ There are some words missing in the last step of the first reduction I think. (Or at least I cannot make sense of the sentence.) $\endgroup$– MaeherCommented Mar 4, 2022 at 10:21
-
$\begingroup$ I've corrected the two sentences of conclusion. $\endgroup$– IevgeniCommented Mar 4, 2022 at 10:36
-
$\begingroup$ I'm talking about "If $b=x$ return a random bit then return $(b- x \mod 3)+1$." I'm guessing the "$=$" should be "$\neq$" and there's an "else" missing, but I'm not quite sure what you were going for. $\endgroup$– MaeherCommented Mar 4, 2022 at 10:38
-
$\begingroup$ okay thanks, I've corrected this step, and changed the notations to make it clearer $\endgroup$– IevgeniCommented Mar 4, 2022 at 13:00
-
$\begingroup$ Could you clarify the second prove? I cannot understand why there is $\frac{1}{3}\mathbb{P}(\mathcal{B}_3wins|b''=2)...$. Do we take $b''=2$ because $m_2=m_b$? $\endgroup$– AwerdeCommented Mar 4, 2022 at 20:00