The prevalence of underpromotion to rook or bishop.

While taking part in the discussion over here

lichess.org/forum/general-chess-discussion/hardestimpossible-composition-challenge

I wondered how often an underpromotion to rook or bishop happens in practical Lichess games, and how often this rare move is meaningful (rather than a joke).

As a test case, I downloaded all 10,680,708 rated standard games played on Lichess in January 2017 (from here: database.lichess.org/) and searched for Bishop promotions using the SCID database, giving me a list of 5035 games which feature a promotion to Bishop. If the SCID search function works correctly at this scale, this means this specific underpromotion happened around once every 2100 games in January 2017.

I am not sure how representative this one month of data is though. In any case, spot checks (including also five games with players rated > 2500) reveal that the move is usually a joke, in cases where the type of promoted piece is totally irrelevant (either because it will be captured in the next move, or because the position is decisive anyway).

Unfortunately, I don't see an easy way to use SCID to filter for more interesting underpromotions, excluding the silly jokes. If someone has an idea, please let me know!

Promoting to rook is useful to protect against accidental stalemate, but I don't know if there's any situation where it's really necessary rather than just a precaution.

@asdf5656
Have a look at the other thread, lichess.org/forum/general-chess-discussion/hardestimpossible-composition-challenge, people there mentioned some unusual and cool examples, including a highly esoteric defensive self-stalemate by not promoting to a queen.

Most of the interesting examples are composed positions though. An exception is this one, which is from a real game from the 1930s:

in my 7000 games across chesscom and lichess i only had to underpromote once, to a knight. i missed it though and ended up losing a drawn game :(

Ok, it's a cold and foggy Sunday morning here, so I scrolled through more than 50 games from my list of 5035 rated Lichess games involving an underpromotion to Bishop. I only looked at classical games, starting at the top in terms of player ratings.

Here's what I found - two interesting positions where underpromotion made sense:

Promotion to bishop is the best move (mate in three if followed up with perfect play), =Q would be stalemate (but another move instead of promotion would win equally fast):


Promotion to bishop is the best (but not the only) move, =Q would be stalemate:


... and some curiosities. Underpromotion is not necessary, but mildly funny and/or beautiful:

Six bishops on the board:


A joke gone wrong - stalemate KBB vs K:


Here, underpromotion seems to signal to the opponent: "I have seen your nasty pin, but I won't give you a queen.":


It does look more beautiful this way:


Quite beautiful:

@Panagrellus Off-Topic but something tells me you will appreciate this game's finale, if you are not yet familiar with it:
www.chessgames.com/perl/chessgame?gid=1238131

@PixelatedParcel
I do like it, thanks!

What's actually much easier to find in the data than the elusive clever underpromotion is the blunder of stalemating - if a player can't resist to promote to a queen at the wrong moment.

Dozens of examples show up if I use SCID on the lichess data and filter for games with the following setting:
- only include games with available computer analysis in the PGN
- promotion to Queen is annotated as a blunder
- the game's result is 1/2 - 1/2.

This nicely shows the power of SCID, but the examples are not very "deep" or interesting. They may serve as a warning though (;

nasty bishop taking away the b7 square:


should have promoted to a knight:


Two queens, but a draw:


Queen, rook and pawn taking away all the squares:


The white king found a safe haven behind a black pawn. Two queens, but a draw:


Unfortunate constellation of Queen and Kings:


The new queen goes to do some work immediately, taking away the g3 square from the black king:


As above, with b7:


Bishops and queen working together to produce an elegant stalemate:
lichess.org/fkJBgyJf#145

I often promote to a rook in a winningendgame, in order to avoid stalemate by blunder. Especially in blitz games, where the opponent refuses to resign. Here is one such game:

If I had promoted to a Queen, it would be a stalemate, so I promoted to a rook. I didn't think about it, I always take rook in these kind of situations. (Yeah, I blundered a Queen few moves earlier, I was just too lazy.)

@LukaCro

That trick doesn't always work though. In the Lichess dataset for January-2017 alone I found roughly 100 games where a promotion to rook caused stalemate (;

Some examples:

I've created two legitimate puzzles where only underpromoting wins. They might arise from real games where no weird stuff is involved. I was hoping they'd stump M. Carlsen but I've had no chance to play him. The computer easily solves all endgames like that because of the lack of hidden information in this game.