Knowing who wins without knowing how — three quarters of a century of 'solving' Hex that keeps establishing a winner without ever handing over the strategy

hexgame-theoryjohn-nashdavid-galestrategy-stealingnon-constructive-proofbrouwer-fixed-pointpspace-completecomputational-complexitytopological-combinatoricsanalog-computingexistence-vs-constructionhistory-of-mathematicspriority-and-attribution

The recurring argument in this cluster is not "Hex connects a lot of fields" — that would be a topic, and high cosine is relatedness, not truth. It is sharper and checkable: across the 1950–1981 results the vault holds about Hex, each has established that the first player wins — or recast the bare fact that someone must — while none of them hands over the how, the actual winning strategy. John Nash proves a win exists and says in the same breath that the strategy "is, as yet, unknown"; his own guess at a construction is left unproven; complexity theory later proves that computing the how is intractable unless P = PSPACE; David Gale recasts the whole no-draw fact as a theorem of topology, strategy still nowhere in sight. The single exception runs the other way and proves the rule: Shannon and Moore's 1950 analog machine produces good moves — pure how — by reading them off a physical field, with no proof of anything at all.

So the map sorts four incommensurable senses of "solving a game" — win by physical equilibrium, win by non-constructive existence proof, win reframed as topological equivalence, and win shown provably hard to compute — against the one gap that runs through all of them: the distance between a solution exists and here is the solution. Hex is the object small enough that all four notions land on the same board, which is exactly why the incommensurability is legible instead of hand-waved.

What makes this a map rather than a list is that the four senses do not agree on what "solved" would even mean, and the disagreement is the finding. Nash's contradiction argument and Gale's Brouwer equivalence are both existence claims that decline to construct; PSPACE-completeness is the theorem that says the construction they skipped is not a historical accident but a wall; and the analog machine is the lone artifact that walks straight through the wall by refusing to prove anything. Cross-linked, not folded, to moc-attribution-and-origin-myths — a second thread runs through the same notes, that each result's authorship is narrower or earlier than its folklore (Nash seven years before "strategy-stealing" had the name; Gale's equivalence was cocktail-party folklore with one direction handed to a colleague).

Titled for the argument — the existence/construction gap — not for "Hex" or "Nash," the most-mentioned names, per the 2026-07-25 lesson.

A winner exists — and the strategy stays unknown

The core of the cluster: three notes on one 1952 document, each showing the same shape — proof of existence with the construction left open, in the discoverer's own words.

• claim-nash-hex-first-player-win-proof-is-non-constructive — the spine. Nash's strategy-stealing argument shows the first player must have a winning strategy (if the second player had one, the first could steal it, since an extra Hex stone never hurts) without ever exhibiting it. The note carries its own repair history honestly: it rested on Tier-4 Wikipedia until an 2026-08-21 pass upgraded the load-bearing leg to Tier 1 (the RAND note below, to Tier 1, plus the PSPACE note in the next section, to Tier 2 — only the non-constructive leg reached Tier 1); it is retained, not deleted, per the vault's append-only correction discipline, and still sits at seedling.
• claim-nash-1952-rand-report-confirms-hex-proof-non-constructive — the Tier-1 anchor, in Nash's own words in RAND report D-1164 (2 Feb 1952): "one can give a simple contradiction argument showing that the player who moves second cannot have a winning strategy and thus that the first player can always win if he plays properly," immediately followed by "The winning strategy is, as yet, unknown." Independently corroborated at Tier 2 by Philip Henderson's 2010 Alberta dissertation. This is the closest the vault gets to watching the existence/construction gap get reported rather than retold — seven years before the argument acquired the name "strategy-stealing."
• claim-nash-1952-dualization-pairing-strategy-conjecture-hex — the guess that didn't close the gap. In the same report Nash speculates, as a guess rather than a proof, that the first player "has a dualization type strategy" built on paired and dummy hexagons. He does not claim it works, and no source read at capture confirms, refutes, or traces the 1952 conjecture forward — the field's real constructive answers came decades later, mostly from computer search. The one place a member note reaches toward how, and it is explicitly unproven.

Why the "how" stays out of reach

The theorem that turns Nash's shrug into a wall: computing the strategy is not merely undiscovered, it is provably hard.

• claim-hex-winner-determination-is-pspace-complete — Even & Tarjan (1976) proved generalized Hex on arbitrary graphs PSPACE-complete; Stefan Reisch (1981) extended it to Hex on the standard board. So an efficient algorithm for arbitrary Hex positions would prove P = PSPACE. Carried honestly on the note and inherited here: the sources prove the decision problem (who wins from a position) PSPACE-complete; the leap to "constructing the winning move is PSPACE-hard" is the field's standard extension, not a verbatim-sourced claim (watch_flag), and Even & Tarjan's primary was read only refracted through Henderson's citation, with the Reisch text via an unattributed mirror. The shape of the story survives those caveats: a 1952 existence proof with no strategy attached, and a 1976–81 pair of results proving, independently, that the decision problem is intractable (no efficient algorithm unless P = PSPACE) — which the field reads, retrospectively, as why no efficient strategy could have been attached. The member note is explicit that this link is "a retrospective one the field draws, not a single continuous argument."

The same fact, recast as topology

Gale's 1979 move is a third sense of "solved": not a strategy and not a hardness result, but a proof that Hex's basic combinatorial fact is a classical theorem of topology — and a candid account of whose idea each half was.

• claim-gale-1979-hex-draw-impossibility-equivalent-to-brouwer-fixed-point — the equivalence itself, Tier 1, quotes recovered directly from the American Mathematical Monthly primary: "Hex cannot end in a draw" is equivalent to the Brouwer fixed-point theorem, not merely analogous — each implies the other. Treated as a founding text of topological combinatorics. Note that this is a claim about that a winner is forced (no draws) recast in another language; the winning strategy is not what is being equated.
• claim-gale-1979-hex-implies-brouwer-via-covering-argument — the direction Gale calls his own (it "only occurred to me recently"): a compactness/covering argument producing approximate fixed points, not an exact construction. Its body was rewritten under a 2026-08-08 cross-model audit (opus-5) that caught the original description matching the wrong proof — a correction preserved in the note's record, and a reminder that this cluster's members are held to primary reads, not paraphrase.
• claim-gale-1979-brouwer-implies-hex-credited-to-stallings-todd — the converse direction, and the attribution tell. Gale states plainly that this half rests "on a suggestion of John Stallings modified by one of Michael Todd," that the equivalence had circulated for years as "cocktail conversation," and that he claims only the n-dimensional generalization as possibly new. The 1979 paper's real originality is narrower than "Gale proved the equivalence" — the half of this cluster that belongs to moc-attribution-and-origin-myths.

The one machine that plays without a proof

The inversion that sharpens the whole map: all how, no why.

• claim-shannon-moore-1950-analog-hex-machine-move-as-saddle-point — Claude Shannon and E. F. Moore's 1950 analog Hex machine set up an electric potential field over the board and read its move off a saddle point in the field — no search, no proof, "won about 70 percent of the games with opening moves." It is the only member that produces actual moves, and it does so by abandoning proof for physics: the equilibrium state is the answer. Where Nash and Gale prove a winner exists without exhibiting a strategy, this machine exhibits playable moves without proving anything about them. Tier 2 — the account is Banerjee (2020) quoting Shannon's own 1953 "Computers and Automata," not the Shannon primary itself.

Entity hubs

Built around this cluster by prior promotions; listed, not built by this pass.

• entity-john-nash — the non-constructive existence proof, in his own 1952 report.
• entity-david-gale — the topology equivalence, and the candid three-person attribution of it.
• entity-stefan-reisch — the PSPACE-completeness of standard-board Hex.
• entity-philip-henderson and entity-ryan-hayward — the 2010 Alberta dissertation (which cites a December 1999 phone call Hayward and Jack van Rijswijck held with Nash) that supplies the Tier-2 corroboration under the Nash notes.
• entity-lej-brouwer and entity-piet-hein — orbit the cluster (Brouwer's theorem; Hein as Hex's earlier independent inventor) though neither is load-bearing in the notes mapped here; Hein in particular appears in none of the eight member notes.

Open threads (honest caveats, not hidden)

• Shannon and E. F. Moore still have no entity hubs — flagged 2026-08-21 (seek-flags.md ~L4147): they built the first Hex-playing machine and are wikilinked from three claim-notes in this cluster, yet neither has a 40-entities/ page. Building those hubs is an entity-hub ask, not this MOC, and is not discharged by this build; it stays open for a future promotion. (I have kept both names as plain text above rather than mint dead wikilinks.)
• Most members are still seedling (Gale's equivalence note is budding). The map records the current footing, not a frozen verdict.
• Three primaries are one read short. Even & Tarjan (1976) is read only through Henderson's citation; Reisch (1981) through an unattributed English mirror (the PSPACE note treats it as supporting texture, not load-bearing — Henderson at Tier 2 clears the floor); Shannon's 1953 "Computers and Automata" through Banerjee's 2020 quotation. None is contested, but none is a first-hand read.
• The "construction is PSPACE-hard" gloss is the field's standard reading of a decision-problem result, not a separately sourced theorem — noted so the map does not harden it.