The Coin in the Middle, and the Table Where It Fails

Take the centre, then mirror every move through it, and you place the last coin. The proof has three requirements and only one of them needs that opening move, which is the step a one-line answer skips. Central symmetry alone is not the condition: an annulus is centrally symmetric and the first player loses on it.

A round table and a pile of identical coins. Two players take turns laying one coin flat on the table, never overlapping another coin and never hanging over the edge. Whoever places the last coin wins. Would you rather move first or second?

Move first. Put your coin exactly in the middle, then answer every coin your opponent plays with its point reflection through that middle. You always have a reply available, so your opponent runs out of room before you do. It works on any table with a centre of symmetry, provided a coin actually fits at the centre. That proviso is not decoration: there is a perfectly ordinary table where it fails and the first player loses.

The route most people take, and why it dies

The instinct is to turn the game into arithmetic: count how many coins fit, check the parity, read off the winner. That instinct is right for any game whose length is fixed in advance.

It cannot work here, and the reason lives in the gap between two words. A maximum packing is one containing as many coins as possible. A maximal packing is one where no further coin fits, which is far weaker: players are free to leave gaps that are individually too small and collectively wasteful. The game ends at a maximal packing, whichever one the play happens to produce, and careless play produces worse ones.

Simulated on a disc whose radius is about 7.7 coin radii, random play against random play ends on totals anywhere between 25 and 32 coins, with both parities appearing. There is no such thing as the number of coins that fit this table, so there is nothing whose parity could decide anything. Any argument that opens by counting has already lost.

The strategy

Point reflection

Put the origin at the table's centre of symmetry. Point reflection sends $p$ to $-p$ . The table is centrally symmetric when $p$ lies on it exactly if $-p$ does, and since $\lvert -p \rvert = \lvert p \rvert$ , a coin centred at $p$ sits wholly on such a table exactly when the coin centred at $-p$ does.

The first player's opening move is the coin centred at the origin. After that he never thinks again: if the opponent plays a coin centred at $p$ , he plays the coin centred at $-p$ . Write $S$ for the set of occupied coin centres. The claim is that this set is symmetric whenever it is the opponent's turn,

S = -S \quad\text{where}\quad -S = \{\,-p : p \in S\,\}

(1)

which holds after the opening move, since the only occupied centre is the origin and $-0 = 0$ . It is preserved by construction. The opponent breaks the symmetry with one coin, and the reply restores it with that coin's mirror.

Fig. 1 — The centre coin is the fixed point of the reflection. Every later coin arrives with a partner on the opposite side.

Why the reply is always legal

The invariant is the easy half. The claim carrying the weight is that the mirror of a legal move is itself legal, and that breaks into three separate requirements.

First, the reply has to fit on the table. That is central symmetry, with nothing further to prove. Second, it has to miss every coin placed before the opponent's move. Immediately before that move the layout was symmetric, so reflection maps the old coins onto the old coins and preserves every distance. If the coin at $p$ cleared them all, the coin at $-p$ clears them all too.

Third, the reply has to miss the coin it is answering, and this is the only requirement that needs the centre coin. It is also the one a one-sentence statement of the method quietly skips. Coins of radius $r$ may not overlap, so their centres are at least $2r$ apart. A coin at $p$ and its mirror at $-p$ are $2\lvert p \rvert$ apart, so

2\lvert p \rvert \;\ge\; 2r \iff \lvert p \rvert \;\ge\; r

(2)

Without a coin in the middle, the opponent can play one straddling the centre, with $\lvert p \rvert < r$ , and the very first reply overlaps the coin it is meant to answer. The strategy would collapse on move two. With the centre coin in place every later coin has to clear it, so

\lvert p \rvert \ge 2r \;\Longrightarrow\; 2\lvert p \rvert \ge 4r > 2r

(3)

and the reply has twice the clearance it needs. That bound is tight rather than generous: on a lattice model of the disc, the closest legal coin centre after the centre coin sits at exactly $2r$ from the middle. Across 300 seeded games against a random opponent, every mirrored reply was tested for legality before being placed, and none was ever illegal.

Why there is a last coin at all

Talking about the last coin presupposes that the game stops, and that needs an argument with no game theory in it. Coins have disjoint interiors and all sit inside the table, so their areas cannot exceed the table's area:

N \pi r^{2} \;\le\; \pi R^{2} \;\Longrightarrow\; N \;\le\; \left(\tfrac{R}{r}\right)^{2}

(4)

For the disc above that caps the game at 59 moves, and the games actually played ran to 35. The cap itself is uninteresting. What matters is that every line of play has bounded length, which licenses backward induction: a game of bounded depth with perfect information and no draws has exactly one player holding a winning strategy. Existence comes free. The mirror argument says which player, and says it constructively.

Over those 300 games the mirror strategy produced totals between 23 and 35 coins, every one odd, which is what it means for the first player to have placed last. The spread is the same spread that killed the counting route, and the strategy is indifferent to it.

The table where the first player loses

Now the qualifier, which is the part usually dropped. It is tempting to state the result as "the first player wins on any centrally symmetric table", and that statement is false. Take an annulus, a disc with a circular hole punched through the middle, the hole comfortably wider than a coin. It is centrally symmetric. It is radially symmetric too, which is a stronger property. And the first player loses on it.

The strategy was never really about the first player. It belongs to whoever can stand at the fixed point of the reflection. On a solid table that is the first player, because he moves before anyone can take the centre away from him. On the annulus nobody can occupy the centre, so player one has to open somewhere off it, and player two inherits the mirror. Requirement three is automatic there, because every coin must clear the hole:

\lvert p \rvert \;\ge\; r_{\text{hole}} + r \;>\; r

(5)

for any hole at all. No coin can cover the centre, so a mirrored reply can never collide with what it answers. Played out on an annulus whose hole has radius three coin radii, 200 seeded games all ended on an even total, so the second player placed the last coin every single time.

Fig. 2 — Same reflection, opposite winner. With the fixed point unavailable, the player who mirrors is the one who moves second.

The condition has a sharper form, worth stating because the annulus can leave you thinking the issue is holes. What the first player needs is for a whole coin to fit centred at the centre, which is more than the centre merely belonging to the table. A centrally symmetric table shaped like a thin cross, arms narrower than a coin, contains its own centre and still refuses the opening move. A table smaller than one coin hands the game over for the dullest reason available: nobody can move, and the player who cannot place is the one who moves first.

What kind of argument this is

This is a pairing strategy. Moves get matched two at a time, and one player guarantees that the second element of every pair stays available to him. Pairing arguments are close relatives of strategy stealing, the trick that shows the first player wins at Hex, but the two deliver different goods. Strategy stealing is non constructive: it proves a winning strategy exists by showing the alternative is absurd, and says nothing about how to play. Here you can execute the proof with your hands.

Notice what the argument never touches. No position is evaluated, no Grundy value is computed, and the space of coin arrangements is never explored. A symmetry of the board did all the work, and the search route it replaces is infinite rather than merely slow.

Sources and further reading

The family this argument belongs to — Strategy-stealing argument
The map $p \mapsto -p$ and its fixed point — Point reflection
Why maximal and maximum differ so much for coins — Circle packing in a circle
The determinacy result behind "exactly one player wins" — Zermelo's theorem
The machinery this problem does not need — Sprague-Grundy theorem

Nothing here was taken on trust. The mirror strategy was played out over 300 seeded games on the disc with every reply checked for legality before it was placed, a further 300 games of random play established that the packing total is not a property of the table, and 200 games on the annulus confirmed that the same reflection there belongs to the second player.