Why do miniature games happen at the grandmaster level?

Question

It is often said that grandmasters calculate many moves in advance before moving a piece. If this is so, why do miniature game (games less than less than 25 moves long) happen?

Answer 1

It is true a GM can think of many moves ahead but sometimes only in particular positions. They also have a vast knowledge of the opening theory like some specific lines in Sicilian or Kings Indian Defence or Ruy Lopez. But there are also limitations. There are some lines which are very sharp and 1 move not played according to the opening book knowledge can pull you down very fast.

Lets say a master of 2200 ELO level is playing the Poison Pawn variation in Sicilian as Black and in the other game he is playing QGD Barmen Variation as Black against a 2400 IM in both games.

Now the opening move order will not matter that much in QGD where some move orders can be changed without affecting the position that much compared to the Poison Pawn in Sicilian. If he makes one callous move in Sicilian Poison Pawn then very soon his game will spiral downwards against an IM who will grab the advantage and squeeze like a python.

Mostly it is the lack of opening knowledge or invention of a novelty which may even backfire due to a lack of understanding of some unfamiliar positions or choosing a wrong plan in middle game.

So there are some reasons why tactical miniatures happen even at Master/GM level.

Answer 2

Your statement, Grandmasters think of next many moves ... is a bit misleading for your question.

Grandmaster are able to calculate many moves ahead, but this is not something they do at each and every move. Basically, as for the start of the game, the first 10 or 20 moves are played from memory (known opening repertoire + home preparation) and not found over the board.

Even when the home preparation ends, players usually do not start calculating like crazy all possible moves to depth 20. They would rather (also see here):

1. evaluate/analyze the position based on general principles and experience
2. depending on this analysis develop a general plan (Should I improve piece activity? Should I play against an opponent's weakness?....)
3. come up with concrete (candidate) moves
4. calculate the consequences of the candidate moves: Depending on the position at hand this can be many moves ahead (e.g. in tactical or forced lines) or it can just be a single move, basically just checking whether you don't blunder something (e.g. in less tactical, more positional games where it is just practically impossible to calculate all possible lines)

So how can grandmasters lose in less than 25 moves?

They are humans so not perfect and might make mistakes. In this case they can go wrong at basically each part of the thought process:

1. They might mix up memorized opening theory, confusing lines and thereby making a bad move. Occasionally they might run into a strong opening novelty prepared by the opponent in which case even playing the commonly accepted opening line can bring them into trouble. Also there is the funny story around the Zapata-Anand miniature where Anand apparently blindly followed published opening theory (informator) of a Miles-Christiansen game which was a prearranged draw but in fact included a huge blunder.
2. They might evaluate the position wrongly, e.g. underestimate a threat.
3. They might come up with a plan that is impossible to push through or pick the wrong plan out of a choice of plans.
4. They might miss a possible candidate move
5. They might make a mistake in calculation.

In addition any combination of the above is possible.

Answer 3

This question can be answered from the angle of computer chess as well, when it might actually be more illustrative. The shortest decisive game ever played in the unofficial World Computer Chess Championship is this 20-move victory by Arasan over AllieStein (TCEC15's round 45). Computers report their depths, e.g. on move 16 AllieStein had a depth/selective depth of 38/54 - indicating it was searching 19+ moves ahead - and yet it was mated four moves later. Why?

The answer is simple: pruning. Some lines a computer (and a human) simply doesn't consider very seriously. For example, take this line:

   [Title ""]
[FEN ""]

 1. e4 d6 2. Qg4

Everyone, including complete beginners, should be able to calculate this far, and yet nobody analyzes the position after 2...Bxg4 for obvious reasons. White is completely lost. Accordingly, we "prune" - we don't consider the lines that arise after this position. If we are White we don't do it because we're not crazy, and if we are Black we don't do it because we assume White isn't crazy (i.e. we only start analyzing it after they play it). Pruning lets us focus our attention on the main moves in the position and analyze those to greater depths.

However sometimes this kind of pruning can miss the key move. Perhaps White isn't actually giving up a queen for free, and there is some deep tactical motif that compensates. We would never find out, because we pruned the move; but perhaps the opponent didn't prune the move, found it was sound, and played it. In that case we can easily lose. Take a look at the position in the aforementioned AllieStein v Arasan game, after 17...Rfb8. (Annotations are by Guy Howarth.)

[Title "AllieStein-Arasan"]
[Event "TCEC15"]
[Date "2019-??-??"]
[Round "45"]
[White "AllieStein"]
[Black "Arasan"]
[Annotator "Guy Howarth"]
[fen ""]
[StartPly "34"]
[StartFlipped "0"]

1.e4 e5 2.Nf3 Nc6 {as mandated} 3.Nc3 Nf6 4.d4 exd4 5.Nxd4 Bb4
6.Nxc6 bxc6 7.Qd4? (7.Bd3) Qe7 8.f3 Bc5 9.Qd3 O-O 10.Bg5 Rb8
11.O-O-O d5 12.exd5 h6 13.Bh4 Ba6 14.Bxf6 Be3+ 15.Kb1?? (15.Rd2) Rxb2+
16.Ka1 gxf6 17.Qxa6 Rfb8 18.Bd3?? (18.Qa5) Qb4 19.Ne2 Rxa2+ 20.Kxa2 Qb2# 0-1

AllieStein projected this principal variation:

18. Bd3 Qc5 19. Qxc6 Qa3 20. Qxf6 Bg5 21. Qd4 Be3 22. Qf6 Bg5

Evaluating it as -2.69. Meanwhile Arasan, when it played 17...Rfb8, projected this principal variation:

17. .. Rfb8 18. Qa5 Qe5 19. f4 Qxf4 20. d6 Bd4 21. dxc7 Rxc2 22. Bb5

When AllieStein played 18. Bd3, Arasan's eval shot up. This indicates it had pruned 18.Bd3 as inferior. Its reply, 18...Qb4, was not the move AllieStein predicted, indicating AllieStein had pruned 18...Qb4 as inferior. But after Arasan played 18...Qb4, AllieStein realized it had blundered, and was mated a few moves later. (Black threatens 19...Qxc3 with nasty discovered checks. If the knight moves, Black has ...Qa3-Qxa2#, ...Rxc2 or ...Bd4, and White cannot defend against everything.)

Something similar to this happens all the time for humans as well.

Technical note: AllieStein is actually a MCTS-based engine, for which the word "pruning" does not apply very well. However, the gist of it remains: AllieStein simply didn't see (or didn't ascribe enough importance to) 18...Qb4, which is why it got mated in spite of searching 38/54 ply ahead.