Hacker News Reader

Comment by hundt | Hacker News Reader

It's because you start with the box labeled (via the initial random labeling) with your name. If you cycle back to it then it means that you found your name on a piece of paper, since the next box you open is always the one matching the piece of paper in the last box. So it is impossible to start in a cycle that doesn't include your name.

View on HN

It assumes, they can secretly assign and remember a 100 items random ordering and then execute it perfectly. Further, it assumes they can decide which orientation is the start vs end of the line. Thus, it's not actually possible, but 'in theory' it seems to work.

EX: If I know your going to order based on which side is closet to the entry door nob when the door is closed. Well nothing says they all enter from the same door if it's based on the wall, put the table in the middle of the room.

I must be missing something. Couldn't one instead assume that they have a private notebook and a pen? And that they can tell left from right? Seems more reasonable.

So you want to add a private notebook that the warden can't read. Further, left vs right does not help if the table is in the middle of a room and people are entering from random doors.

AKA, the solution assumes a lot of information not explicitly part of the original wording.

I agree that the puzzle sort of plays dirty pool here.

The strict requirement is that the boxes are distinguishable. The puzzle implies this by saying they're lined up on a table, but it would be more honest to the reader to say they're numbered 1-100, or somesuch.

Couldn't I start in a "tail" leading to a cycle that doesn't include my name? Suppose my name is "A", and I proceed:

check box A - read name B; check box B - read name C; check box C - read name B

and now I'm in a B-C loop, and will never find my name.

Re-read your scenario -- you have both box A and box C containing name B, but the problem statement says each name appears only once. So you can't get into cycles like this.

Ah, of course! That guarantees there are no "tails"; every name (& box) is part of exactly one cycle. And this is then a property of cycles of permutations generally. Thanks!

No, because in your example both boxes A and C have name B in them. That's against the rules.

No, then "B" would be in two boxes: A and C.

I don't understand: the solution text claims that some permutations are guaranteed to work every time[1]. But you still have to initially land in one of "your" cycles, right?

[1] "If it happens that the permutation has no cycles of length greater than 50, this process will work every time and the prisoners will be spared."

You always land in "your" cycle, because you start with the box with your name on it.

All sequences of box-opening that use the method described must eventually cycle, because both box-to-name mappings are 1-to-1. Because it's a cycle, the sequence must eventually lead back to the box you started with. Since you start with the box with your name on it, then whatever cycle you landed it, it definitely contains the box with your name on it, because that box is the one that completes the cycle. The only question is whether the cycle is length-50 or less.

I guess that's what I'm confused about: when we start, the boxes we map our names to have nothing to do with the name in the box. Just because my cycle includes the box I got assigned to, doesn't mean that cycle includes the box with my name inside it. If my name maps to 89, then sure, I accept 89 is a LT-50 cycle, but why does it mean that that cycle actually contains my name (as opposed to the box we assigned me to)?

Think about how you will get back to the box you got assigned to. Like what do you have to see in order to return to box 89, where you started?

Ah, okay, that makes more sense. So the solution is to impose a random ordering on the (1-100 random numbers assigned to the) boxes that ensures you will return to the first one you picked, and that the one that points to your initial pick (or your initial pick itself) has your name in it.

This ordering then partitions the boxes into cycles where each prisoner is in their cycle, and has a low chance of having 50-sized cycles.

Still mesmerized at how the choice of query can improve your chances like that without accumulating information.

But why would that matter? You don't know 89 has your name in it, because the numbers were randomly assigned to random people in the first place. You would just draw the name associated with box 89, which is not necessarily your name.

Whatever box you open, you take the name inside and go to the number you associated with that name. So when you end up at box 89, the last box you opened must have had your name inside. So you will have found your name, provided you got there in under 50 steps.

Do you have any intuition to why it helps to build the query sequence on the data? It seems natural that if you have to plan the queries for everybody in advance, you can't do better than 2^(-100) or perhaps (50!)^2/100!. Yet somehow by using the return values to find our next query point we can do much better. Blows my mind.

Each individual still has a ~50% chance of being unable to find their name using the query sequence protocol. By agreeing to the same query sequence their success modes and failure modes are now linked. If there's a cycle of 51 or greater, all of those individuals in that cycle are guaranteed to fail, while if they're in a cycle of 50 or less, all of those individuals are guaranteed to succeed.

What this protocol does is make it so each individual failure mode is much more likely to overlap with each other failure mode. Correlating the performance of individuals when you have a one fail all fail scenario is a common solution to this type of problem.

You are right, but we can also agree on a common query sequence in advance that doesn't use the 'results' of the queries. What about using what we find makes this so much more efficient?

You need to use the result of the query to place yourself in the right cycle (in this case by starting with the box matching your name).

Ah right, that's clever. Thanks for the explanation!

Reply on news.ycombinator.com