Answer:
Seat the six pirates at random around the table, and fix attention only on pirates with ratings and (since can never be eliminated). At any moment, list the remaining pirates clockwise as , where is the number of surviving pirates strictly between and clockwise from to , and is the number strictly between and on the other arc. Let be the probability that the final duel is between and given this configuration. It is clear that this probability depends only on the numbers and , not on the specific identities or ranks of the pirates and . Indeed, every and has rank at most , so any duel involving or eliminates the weaker pirate regardless of which or it is, while duels among the ’s (respectively among the ’s) merely reduce the size of that block by one.
If or , then and are adjacent with pirates left. In that situation, at each turn exactly one choice of challenger (namely or , depending on who is on the right) forces the duel vs immediately and eliminates ; all other choices remove someone else. Hence the probability that survives until the last duel telescopes as
|
Assume now , and keep for the number of remaining pirates. Consider the clockwise block of pirates starting at and stopping at the pirate immediately before ; this block has exactly members, namely . If the randomly chosen challenger lies in this block, then the ensuing duel occurs within this block (the challenged pirate is the next clockwise pirate, still inside the block, except for who challenges ), and in every case the eliminated pirate is one of . Thus choosing the challenger from this block decreases by . By the same reasoning, the clockwise block starting at and stopping immediately before has members, and choosing the challenger from that block decreases by . Since the choice of challenger is uniform among the remaining pirates, we obtain, for ,
|
Initially . Using the boundary values , , , and symmetry , the recurrence gives , , , and . Since the initial seating is uniform, the clockwise gap between and is uniformly distributed on , so the sought probability is
|
Alternative solution. We will solve the problem in a more general form for pirates instead of six. Let denote the probability that the last two surviving pirates are and . Our goal is to prove the recurrent formula
|
valid for .
Consider the first step of the ritual. Pirate is eliminated immediately with probability : indeed, the probability that and sit next to each other is , since if the pirates are seated starting from pirate , then pirate has possible places to choose from, and exactly of them are next to . Once they are adjacent, the duel between and happens with probability . Hence the probability is
|
Thus, with probability , the first eliminated pirate is one of . In that case, the remaining pirates are again seated around the table in a random circular order, with all such orders equally likely: if pirate is eliminated, then any given resulting circular order can arise in exactly ways, since must have sat next to and dueled one of the stronger pirates , either by challenging or by being challenged by . This number depends only on and not on the resulting order, so after relabeling the survivors as , we obtain the original problem with pirates. This proves the recurrence .
It remains to compute . Since , repeated use of gives
|
Therefore, the probability that the final duel is between pirates and is . More generally, one can also prove by induction from that