Skip to content
Solved Games

Losing chess

Chess & chess variants · all families · data

Also known as: Antichess, Suicide chess, Giveaway chess, Killer chess, Take-all chess, Capture chess.

White (first player) wins; 1.e3 forces a win against every Black defense
Result First player wins
Strength Weak
Method Proof-number search, Retrograde analysis, Alpha-beta + DB
Year 2016
Solved by Mark Watkins (University of Sydney, School of Mathematics and Statistics)
Source Watkins 2016
Board Standard 8x8 chessboard, orthodox starting position; captures compulsory, non-royal king, no castling, promotion to king allowed
Distance to result Watkins’ proof has a maximal upper-level (best-first) tree depth of 94 ply, with endgame tablebases adding further plies; the hardest defenses originally ran very deep (e.g. 1.e3 e6 ~197 moves). A later refinement (tolius, 2021) shows White wins in under 80 moves with the 50-move rule applied (e.g. 1.e3 c5 mates in #79).

Solution history

Section titled “Solution history”
Date Size or opening Result Strength Credit Source
2016 standard game first player wins Weak Mark Watkins source

One row, derived from the record’s year, solver and primary source; no fuller history is recorded yet.

References

Section titled “References”

Resources

Section titled “Resources”

Tablebases

Code

Write-ups

Notes

Section titled “Notes”

Losing chess (antichess) on the standard 8x8 board with the orthodox starting position: captures are compulsory (the side to move chooses among multiple captures), the king is a non-royal piece (no check/checkmate, promotion to king is allowed), and castling is illegal. A player wins by losing all their pieces or being stalemated (under the stalemate ruleset). Weakly solved by Mark Watkins in October 2016 (announced Oct 10, 2016; peer-reviewed in ICGA Journal 39/2, 2017): the first move 1.e3 is a forced win for White against any Black reply. This is a WEAK solution — a winning strategy is proven from the start position (a proof tree for 1.e3), NOT a full tablebase giving the value of every reachable position, so it is not “strong.” Twelve of Black’s 20 replies had been known lost for ~two decades; Ben Nye solved a 13th in 2003; Watkins (with contributions from Klaas Steenhuis, Catalin Francu/Nilatac, Stan Goldovski, Fabrice Liardet and others) closed the remaining lines, the last being 1.e3 b6 (Liardet Defense). Method: proof-number search (best-first pn-search as an evaluation function feeding an upper-level proof tree) combined with custom Losing Chess endgame tablebases (4-, 5-, and 6-unit, distance-to-conversion metric, built via retrograde analysis). The final proof tree (graph) has ~863,301,867 nodes (~900 million positions) with ~66 million transposition pointers; Watkins estimates roughly 10^16 total positions searched over the project (~100x the checkers effort). Result holds under “joint FICS/International Rules” for stalemate (a win under joint rules is a win under both FICS and International stalemate conventions); does NOT cover the vinciperdi ruleset (stalemate-with-pieces is a draw there). Watkins gives no single citable state-space or game-tree complexity figure for the full game (he notes orthodox-chess estimates of ~10^40 legal positions and ~10^83 game-tree complexity as a loose comparison, adjusted up/down by the forced-capture rule and non-royal kings), so both complexity fields are left null. Distance: original proof had a maximal upper-level tree depth of 94 ply (tablebases add more), with some defenses running very deep; a 2021 refinement by “tolius” (lichess) reduced the longest forced win to under 80 moves with the 50-move rule applied (1.e3 c5 mates in #79). Independent confirmation: Klaas Steenhuis and Lajeunesse transferred/verified proofs to FICS rules with an independent searcher, and the later tolius refinement re-derived and improved the result with different tooling.