Typical arguments for the longest possible game of chess take a form something like this.

1) Each player moves their knights back of forth 49 times 2) Each player moves one pawn forward one square and repeats the knight shuffling. 3) Once all pawns are on the 4th and 5th ranks, .... and so on.

I,we found estimates around 9000. I'm interested in the longest possible game of chess, with the restriction that the players are actually trying to win, and not cooperating to make the longest possible game allowable by the rules. We might call this the longest realistic game of chess.

Since the above argument totally breaks down, what are some new considerations that might allow us to lower the upper bound on the longest realistic game of chess?

  • It's hard to combine a rigorous requirement like longest possible with the vague notion of "trying to win". Could you try to define what "trying to win" would be in a more formal way, or alternatively are you just looking for the record so far? Nov 8, 2019 at 8:27
  • Trying to win is well defined. Try to find the best move in every position. Formally, best play is given by the minimax algorithm.
    – psitae
    Nov 8, 2019 at 9:42
  • I'm not looking for the record. I to reformulate the type of argument seen above where cooperation is not allowed. Sorry this wasn't clear.
    – psitae
    Nov 8, 2019 at 9:43
  • It is pretty clear that in many situations competently trying to win, will shorten the game compared to incompetently trying to win. But if you make the players too incompetent, in which way can you still meaningful say that they are trying to win? You end up with random moves and the longest random game is one that is indistinguishable from a collaborative effort. Nov 8, 2019 at 10:00
  • The min-max algorithm depends on a value-function. The value-function tells you how "good" a position is. How good a position is, is not objective. It's relative to your opponent or your pool of opponents. "Best play" is not formally defined except maybe in winning positions where there is a shortest way to mate against best defense. Nov 8, 2019 at 10:05

2 Answers 2


The longest game I've ever found in tournament play is Nikolic - Arsovic, played in Belgrade in 1989. You can watch it here. It ended in a draw after 269 moves in a rook+bishop-vs-rook endgame

  • 1
    This is a good sanity check for any sort of result from logic, but this is not an answer to my question.
    – psitae
    Nov 8, 2019 at 9:43
  • Then please clarify your question. What does "non-cooperative" mean?
    – David
    Nov 9, 2019 at 10:13

If, in some chess position, one side can win against any defence, then we may ask what the least n, where the winning side can guarantee to checkmate in n moves or fewer, against any defence.

Analysis of seven-piece positions has brought to light some positions where n is as large as 549, for example this one in KQPKRBN.

Presumably if the number of units allowed on the board were increased, there'd be positions whose shortest wins are even longer.

But for a game to reach such a position, one side must have already blundered a half-point (supposing that the initial position is a draw with best play). So long as neither player blunders into a position where the other player can force a win, how do you decide which possible moves are co-operative and which are not?

Not the answer you're looking for? Browse other questions tagged or ask your own question.