THE HIDDEN MAP OF CHESS OPENINGS
Chess players learn openings by name: the Sicilian, the French, the Ruy Lopez, the King’s Indian. These names are useful for talking about chess, but they are a historical patchwork. One broad label, such as the Sicilian Defence, covers structurally very different systems; other names designate a single precise move sequence; and different move orders can land on the very same position. As early as 1925, a chess course by Réti questioned the scientific basis of opening theory.
Marc Barthelemy, of Université Paris-Saclay, and Manlio De Domenico, of the University of Padua — two physicists who study complex systems — propose a different map, built from positions rather than names.
Slow pawns, fast pieces
Their starting point is a rule every beginner knows: pawns cannot move backwards. Each pawn advance is an irreversible decision that reshapes the board for every other piece. Knights, bishops, rooks and the queen, by contrast, can be rerouted in a few moves.
To a physicist, that is a familiar situation: a slow part of the system that keeps a memory of the past, and a fast part that reorganizes on top of it. The pawn structure becomes the slow skeleton of the position; the pieces, its fast degrees of freedom.
From 3,810 names to 619 positions
The authors use an open catalogue published by Lichess, listing 3,810 named variations. These are definitions from opening theory, not games actually played: no results, ratings or engine evaluations enter the study. Every line is cut after ten half-moves — five moves each — the depth at which the number of distinct positions is close to its maximum.
At that depth, 1,749 lines remain; the others are defined by fewer than ten half-moves. They reduce to 682 move sequences, mostly because many names share the same first ten half-moves, and then to 619 distinct positions, often because the same moves were played in a different order. Those positions carry only 435 distinct arrangements of pawns for both sides, so each arrangement is shared on average by 1.42 positions: the pawns sum up much of a position, but not all of it.
Measuring an opening
For the slow part, the team counts pawn commitment: the total number of rows a side’s pawns have advanced, including pawns that were later captured. It ranges from 2 in some Four Knights and Italian lines to 9 in the Four Pawns Attack against the Alekhine Defence. They also note where surviving pawns stand — centre, c-file or wings — and whether opposing pawns block each other or threaten captures. Of the 133 positions with two or more such contacts between pawns, 119 mix both kinds.
For the fast part, they measure how many squares the pieces attack and how strongly that activity points to the centre. The two sides of the picture turn out to be largely independent: for White, pawn commitment and piece centralization correlate at only 0.32, and for Black at 0.06 (on a scale where 0 means no link and 1 a perfect one).
One continent of pawn structures
The most striking result comes from a network. Each node is one of the 149 different White pawn structures seen after five moves. Two nodes are linked when a single pawn move, or the loss of a single pawn, turns one into the other. All 335 links point one way, since pawns never go back and captured pawns never return.
139 of the 149 structures form one connected network, carrying 608 of the 619 positions — 98%. The ten isolated structures are early configurations from lines such as the Portuguese Gambit, the Grob or the Hippopotamus. Despite the variety of names, almost every opening sits on the same continent.
Eight regions
An algorithm that looks for groups of densely linked nodes finds eight communities. Random networks with the same number of links per node produce about as many groups, but much less sharply separated: the real network’s modularity — a score of how cleanly a network splits into groups — is 0.569, against 0.444 for the random ones, a gap about 17.5 times the typical spread among those random networks.
The communities were found without using any opening names, yet their central structures are familiar. One gathers the fianchetto set-ups — a bishop developed behind a pawn pushed one square — typical of English and Catalan systems; others belong to the queen’s-pawn family from which the Queen’s Gambit and the Slav develop; another, to a broad White pawn centre. The authors warn that these are “early structural families”, not exact matches with the classic pawn formations of the middlegame.
Choosing by profile
The team turns its measurements into scorecards for individual positions. The King’s Gambit line studied is the most committed of six examples, with little piece activity; the Giuoco Pianissimo shows almost the opposite, minimal pawn commitment and pieces close to the most active in the catalogue. The analysis does not yet say how often openings are played; a database of real games could show which regions players actually visit.
The proposal for players flips the usual habit. Instead of choosing an opening family by name and studying its lines, a player could describe the positions they like — few pawn commitments, central pieces, unbalanced structures, little theory to memorize — and look for whatever openings lie close to that profile.
