Lufting or not lufting
Contents
You have castled, the game is quiet, and your hand drifts towards the h-pawn. Good idea or wasted tempo? We had an engine analyse 18,234 pushes taken from 47,852 amateur games. Verdict: h3 is neither good nor bad in itself. Five questions are enough to decide; under each rule, ten real positions to practise on.
The essentials
Luft, German for “air”, gives the castled king an escape square. The rules below are statistical: they come from the measured quality of thousands of moves actually played, and say which choice goes wrong least often in a given situation. They give a quick answer to apply by default. They do not replace analysing the position: they are trends, not absolute truths, and every set of exercises contains exceptions.
Rule 1: opposite-side castling, do not push
When the kings are castled on opposite sides, each player throws their pawns at the enemy king. The advanced pawn no longer protects anything: it gives the attack something to hook onto (Coach-Bill, 2014). Here h3 is a mistake 29.5% of the time. Pushing two squares does not help: h4 is a mistake 38% of the time, including when the enemy pawns are already coming.
The hook
Without a white pawn on h3, a black pawn arriving on g4 has nothing to attack. With a pawn on h3, it strikes it, and a line opens in front of the white king.
The example, with Black: 8...h6 and 9.g4
White king on c1, black king on g8. After 8...h6, 9.g4 launches the white g-pawn: 9...Nc6 10.Bxf6 Bxf6 11.g5, and the pawn strikes h6; Black’s shelter opens up. 8...Nc6, a simple developing move, was worth 1.7 pawns more.
Your turn: ten positions
All with the kings castled on opposite sides. Most often the push is a mistake; a few exceptions are a reminder that the rule gives a trend.
Rule 2: if the bishop can take without a recapture, do not push
The mechanism: you push h3, an enemy bishop takes on h3, and your g-pawn cannot recapture, because the enemy queen, on the g-file, pins it to your king (Gserper, Storm). In the 22 positions measured where this mechanism existed, the push was a mistake 14 times (64%). That is few cases, but the mechanism can be checked move by move on the board. This rule comes before rule 3, even when an enemy bishop or knight threatens to pin.
With White: 6.h3? Bxh3
The black queen on g6 faces the g2 pawn and the king on g1. The black bishop on c8 sees h3: the d7, e6, f5 and g4 squares are empty. After 6.h3, 6...Bxh3: the pinned g2 pawn cannot recapture. 6.h3 cost at least 2.5 pawns.
With Black: 19...h6? Bxh6
The white queen on g4 faces the g7 pawn and the king on g8. The white bishop on e3 sees h6: the f4 and g5 squares are empty. In the game, 19...h6 was met by 20.Bxh6: the pinned g7 pawn could not recapture. 19...h6 cost at least 1.4 pawns.
Your turn: ten positions
In each one, after the push, an enemy bishop can take on the pawn’s square without your g-pawn, pinned by the queen, being able to recapture. Most often you should not push; a few positions are exceptions, when the capture gains nothing.
Rule 3: against a pin, push, and early
This is the classical use of h3: taking away the square from which the enemy bishop would pin your knight to the queen (Gserper, 2012 (1)) (Gserper, 2012 (2)). Timing matters: the push is a mistake only 7.5% of the time before move 9, but 19% of the time after move 16, as often as an ordinary move.
Timing matters
Share of h3 and h6 pushes against a possible pin that are a mistake, by the moment they are played.
The example, with White: the Closed Ruy Lopez
White wants to play d4, but the black bishop on c8 would come to g4 and pin the white knight on f3, which defends d4. 9.h3 first takes the square away from the bishop: it is the theory move and the engine’s first choice.
The trap: check the reply
Here the black bishop on g4 already pins the white knight on f3 to the queen on e2. But 10.h3 allows 10...Nd4: the black knight attacks the queen, and the pinned f3 knight cannot take it without giving up the queen. White loses almost three pawns; 10.c3, which covers d4, kept the balance.
Your turn: ten positions
In each one, an enemy bishop or knight can come to g4 (g5 when you have Black), or a bishop is already there, and the kings are not castled on opposite sides. Look at the move number too.
Rule 4: without a reason, do not push
If no enemy bishop or knight can come to g4, and no enemy bishop is already there, h3 prevents no pin. It is a mistake 20% of the time, slightly more than an ordinary move (18.7%), and the rate changes little with the stage of the game: 16% before move 9, 20 to 21% afterwards. Chasing away an enemy knight or queen already on g4 belongs here too: 20.4% mistakes. An early h3 does not develop any piece (Illingworth): the danger is the move you did not play instead.
With White, a tension to resolve: 11.exf5
The white e4 pawn and the black f5 pawn face each other. 11.exf5 resolves the tension; the game went 11.h3, and Black used the time given to take the initiative in the centre. Cost: about one pawn.
With White, a threat to meet: 15.Nxe4
The black knight on e4 threatens to go to d2 and attack the white queen on f3 and the rook on f1. 15.h3 allows that fork and loses the exchange; the knight had to be removed.
Your turn: ten positions
In each one, no enemy piece is on g4 (g5 when you have Black) and no enemy bishop or knight can reach it. Work out whether the push still has a reason to be played.
Rule 5: if you push, prefer the h-pawn
Every advanced pawn weakens squares, and not the same ones. h3 takes its protection away from only one square; g3 leaves two without a pawn to defend them, f3 opens a diagonal towards the king. As a result, g3, f3 and h4 are mistakes almost three times in ten, twice as often as h3, whatever the stage of the game.
Which push costs the most?
Share of moves that are a mistake, for each push of the kingside shelter. The figures combine White and Black: “h3 / h6” counts h3 played by White and h6 played by Black.
h3: one square weakened
With White, the king gains the h2 square and only g3 loses a defender. With Black, it is the same in mirror image: after h6, the king gains h7, and g6 is weakened.
g3: two squares weakened
With White, f3 and h3 lose their defender; with Black, after g6, it is f6 and h6. Without the bishop that travels on those squares, the holes stay for the rest of the game.
f3: an open diagonal
With White, the f2 square empties: the diagonal leading to the king on g1 opens, and e3 and g3 are weakened. With Black, after f6, it is f7 that empties, and e6 and g6 that are weakened.
Your turn: ten positions
Pushes of the g-pawn, the f-pawn, or the h-pawn to h4 (h5 for Black), all with the kings castled on the same side. Most are mistakes; three are justified.
Where these rules come from
We went back over 47,852 amateur games analysed by ChessPivot (Lichess and Chess.com, 400 to 2399 ELO, bullet excluded) and picked out every pawn pushed in front of a castled king before move 26. For 18,234 of them, Stockfish 18 compared the move played with the best move, in the same search. A loss of at least one pawn counts as a mistake. Benchmark: 8,000 ordinary moves drawn from the same games.
And the back-rank mate?
It is the most quoted argument for luft (Naroditsky, 2015). Among 57,409 sides that castled kingside, those that had pushed a shelter pawn were mated on the back rank slightly less often below 1600 ELO, not above. Luft against the back rank belongs later, when the rooks are active.
| Side’s rating | With luft | Without luft |
|---|---|---|
| 400-799 ELO | 5.3% | 6.3% |
| 800-1199 ELO | 3.8% | 4.2% |
| 1200-1599 ELO | 3.0% | 3.2% |
| 1600-1999 ELO | 2.2% | 2.0% |
| 2000-2399 ELO | 1.9% | 1.2% |
What the measurement does not say
The rates cover all moves, not a per-player average. The context is recorded with simple rules: that a piece can come to g4 does not mean it threatens anything there. Comparisons between games are associations, not effects.
In your games
Go back over your last ten games in which you castled. For every pawn you pushed in front of your king, ask the five questions in order: opposite-side castling, a capture with no recapture, possible pin, a reason to push, the pawn chosen. Note the pushes played for no reason and what you could have played instead.
Going further: the sources cited
The calls in brackets lead to the entries below. Every measured figure comes from our own analysis, described in “Where these rules come from”.