# Difference between revisions of "Open Problems:79"

Joshuabrody (talk | contribs) |
(Formatting) |
||

Line 3: | Line 3: | ||

|who=Joshua Brody | |who=Joshua Brody | ||

}} | }} | ||

− | + | Cryptogenography, introduced by Brody et al. {{cite|BrodyJSW-14}}, is concerned with the following question: “How to share a secret without revealing the secret owner?” | |

− | Cryptogenography is concerned with the following question | + | In this problem, there are $k$ players and an eavesdropper. Input is a random bit $X \in \{0,1\}$, also called “the secret.” The secret owner $J$ is chosen uniformly at random from $[k]$. Players have private randomness and they can communicate publicly on a shared blackboard visible to everyone. Players are said to win if |

− | |||

− | In this problem, there are $k$ players and an eavesdropper. Input is a random bit $X \in \{0,1\}$, also called | ||

* everyone learns the secret, and | * everyone learns the secret, and | ||

* eavesdropper does not guess the secret owner correctly. | * eavesdropper does not guess the secret owner correctly. | ||

− | We are interested in maximizing the probability that players win (maximum success probability). | + | We are interested in maximizing the probability that players win (maximum success probability). For $k=2$, the following trivial protocol has success probability $0.25$. Just output $1$. It can be shown easily that maximum success probability is at most $0.5$. The bounds can be improved to $0.3384 \le $ maximum success probability $\le 0.361$. Here is a protocol that achieves success probability $1/3$. |

− | |||

− | For $k=2$, the following trivial protocol has success probability $0.25$. Just output $1$. It can be shown easily that maximum success probability is at most $0.5$. The bounds can be improved to $0.3384 \le $ maximum success probability $\le 0.361$. Here is a protocol that achieves success probability $1/3$. | ||

'''First round:''' | '''First round:''' | ||

Line 27: | Line 23: | ||

If the speaker has the secret, announce it, else announce a random bit. | If the speaker has the secret, announce it, else announce a random bit. | ||

− | + | For large $k$, the following bounds can be shown: $0.5644 \le $ maximum success probability $\le 0.75$. Can we improve these bounds? For more information, including state of the art bounds, see: {{cite|Jakobsen-14|DoerrK-16}} | |

− | |||

− | For large $k$, the following bounds can be shown: $0.5644 \le $ maximum success probability $\le 0.75$. | ||

− | |||

− | |||

− | Can we improve these bounds? | ||

− | |||

− | For more information, including state of the art bounds, see: {{cite|Jakobsen-14|DoerrK-16}} | ||

− |

## Revision as of 03:16, 28 April 2017

Suggested by | Joshua Brody |
---|---|

Source | Banff 2017 |

Short link | https://sublinear.info/79 |

Cryptogenography, introduced by Brody et al. [BrodyJSW-14], is concerned with the following question: “How to share a secret without revealing the secret owner?” In this problem, there are $k$ players and an eavesdropper. Input is a random bit $X \in \{0,1\}$, also called “the secret.” The secret owner $J$ is chosen uniformly at random from $[k]$. Players have private randomness and they can communicate publicly on a shared blackboard visible to everyone. Players are said to win if

- everyone learns the secret, and
- eavesdropper does not guess the secret owner correctly.

We are interested in maximizing the probability that players win (maximum success probability). For $k=2$, the following trivial protocol has success probability $0.25$. Just output $1$. It can be shown easily that maximum success probability is at most $0.5$. The bounds can be improved to $0.3384 \le $ maximum success probability $\le 0.361$. Here is a protocol that achieves success probability $1/3$.

**First round:**
If Alice has the secret,

- with probability $2/3$ she asks Bob to speak second
- with probability $1/3$ she speaks second.

Else (vice-versa)

- with probability $2/3$ she speaks second
- with probability $1/3$ she asks Bob to speak second.

**Second round:**
If the speaker has the secret, announce it, else announce a random bit.

For large $k$, the following bounds can be shown: $0.5644 \le $ maximum success probability $\le 0.75$. Can we improve these bounds? For more information, including state of the art bounds, see: [Jakobsen-14,DoerrK-16]