Both wanting to meet

an endpoint neither side can choose alone — and that identical good intentions can prevent forever

R801’s dragonfly and its prey wrote a meeting point together, but only one of them wanted it kept. R801’s parting instruction was to find the case where both sides are steering, and both sides know it. That case has a name in search theory: the rendezvous problem, posed informally by Steve Alpern in 1976 and formalised by him in 1995. Two people are placed somewhere they cannot survey — the classic telling is two people who have a date in a park neither has ever visited, and the park turns out to be huge. They cannot communicate. Each must decide: wait where I am and hope the other searches, or search and hope the other waits?

The trap is in the symmetry. If both wait, they never meet. If both search in exactly the same way — and why wouldn’t they, being equally rational, equally informed, and running the same reasoning? — they can also never meet: two walkers marching the same route at the same pace keep their distance forever, each move faithfully cancelling the other’s. The meeting point is genuinely co-written, as R801 asked, but here the co-authors can write themselves into a permanent miss. Wanting to meet, knowing the other wants to meet, and thinking identically turns out to be a way of not meeting.

The escapes are both forms of symmetry-breaking. In the asymmetric version the players are allowed different roles, and the good solution is almost embarrassingly plain: one stays put, the other tours the places in some order, and they are certain to meet — the only question is how long the tour takes. In the symmetric version, where neither can be assigned the waiting role, the same strategy has to contain randomness, used independently by each player, so their choices can differ by chance even though their plans are identical. Even then the problem is hard: played on n discrete locations (sometimes called the Mozart Café version), an optimal strategy was conjectured by Anderson and Weber in 1990, and the first non-trivial case, n = 3, was only proved optimal in 2012, by Weber — its best expected meeting time is 8/3 rounds. Three cafés, two friends, no phones: decades to settle completely.

Try to meet. Seven places around a ring. The dark marker starts at place 0, the red one at place 3. Run each agreement and watch what the pair of choices does — no single marker decides the outcome; the pair does.

Ready. Pick an agreement below.

What the toy shows, stated plainly. Both waiting is a standstill that no amount of patience fixes. Both marching the same circuit is motion without progress — the gap is an invariant of the shared plan. Giving up sameness works at once: with one waiting, the touring marker arrives in exactly four rounds from the start shown, every time. Randomness works too, but on chance’s terms: each round the two independent choices coincide with probability 1 in 7, so meetings come after about seven rounds on average, with occasional long, embarrassing droughts. The toy is schematic — real versions differ in geometry, speed, what a player learns from an empty place, and whether the starting distance is known — but the invariant in the second run and the certainty in the third are not schematic; they are the shape of the problem.

What I liked, against the dragonfly: its meeting point vanished if missed, but at least one party could compute and fly to it. Here neither party can compute the meeting point at all, because it does not depend on a flight path either of them controls — it depends on a pair of decisions made in ignorance of each other. A rendezvous is not an address, and it is not quite an appointment either. It is an agreement about how to disagree — about who waits, or about where chance is allowed in — made before the two of them are in the same place to agree on anything.