A|H pawn && wrong bishop v K

[op12no2]

But you have the right bishop in that scenario - if you change it to black it's a draw according to Hiarcs endgame database...

Edit - Ah you changed it. Yes Hiarcs says white win now. Bugger! Need to add white king position into it.

Thank you.

[hgm]

Sorry, I was still editing it. Fruit sees a mate in this position.

[op12no2]

Yes, you're right. Not as simple as I thought...

[Ferdy]

Here is my basic heuristics used in Deuterium. The defending king should be 0 or 1 unit distance away from the promote square.
The promote sq is at a8. The black king in the diag, should be either in a7 or a8 or b7 or b8. When black king is at a8 sq, the distance is 0 otherwise its distance to a8 is 1, i.e if black king is a7, b7, b8.
[d]8/k7/8/P1K5/5B2/8/8/8 w - - 0 1

[hgm]

Indeed, this is what I do now too. But it is a bit conservative; there are many positions that could be recognized earlier. Namely where the black King would arrive at the promotion square 3 ply before the white King and 3 ply before the Pawn.

[Ferdy]

Agreed there are things that someone can add when the defending king is not close to that promotion sq. This is why I call it basic for quick implementation.

[hgm]

What positions do you recognize as draw in KQKP? This is one I haven't implemented yet. I suppose it is basically all positions where the defending King covers the promotion square, and for c/f Pawns also where it is in the corner, as long as the Queen cannot step in front of the Pawn. I guess you need that one only when the Queen is a Knight jump away from the corner.

The tricky thing is to know how close the attacking King can be before it is lost. I suppose that for c/f Pawns it matters if you are defending on the short or long side of the Pawn, as on the long side you would be drive in front of the Pawn at least once, giving him the opportunity to get one step closer.

[kbhearn]

What positions do you recognize as draw in KQKP? This is one I haven't implemented yet. I suppose it is basically all positions where the defending King covers the promotion square, and for c/f Pawns also where it is in the corner, as long as the Queen cannot step in front of the Pawn. I guess you need that one only when the Queen is a Knight jump away from the corner.

The tricky thing is to know how close the attacking King can be before it is lost. I suppose that for c/f Pawns it matters if you are defending on the short or long side of the Pawn, as on the long side you would be drive in front of the Pawn at least once, giving him the opportunity to get one step closer.

Depends what % correct you want to be. if you can accept both some false-draws and false-wins that take a couple ply of search to resolve correctly the rules for that are pretty well codified in books. i.e. the win zone for an a-pawn with white to move:

[d]5Q2/8/8/KKKK4/KKKKK3/3KK3/pk1KK3/3KK3 w - - 0 1

the problem would of course be that there are mild exceptions to the rule where the white king can enter the zone with a discovered check and win

The two similar win zones for the c-pawn (depending on defending king short-side or long-side)

[d]5Q2/8/8/8/KKK5/3KK3/1kpKK3/4K3 w - - 0 1

[d]5Q2/8/8/KKKK4/KKKKKKK1/KK3KK1/KKpk1KK1/K4KK1 w - - 0 1

[op12no2]

HGM: Indeed, this is what I do now too. But it is a bit conservative; there are many positions that could be recognized earlier. Namely where the black King would arrive at the promotion square 3 ply before the white King and 3 ply before the Pawn.

Thanks, I narrowed it down as per your and Ferdinand's rules and it evaluates OK in testing, e.g. this evaluates to 0, as does moving the WK to B1|B2|A2.

[d]8/8/8/3b4/p7/8/8/K6k w - - 0 1

But analyse from there I get:-

Code: Select all

4.31 22/27 (-244cp) Kb2 Kg2 Ka3 Bb3 Kb2 Kf3 Ka3 Ke4 Kb2 Kd3 Ka1 a3 Kb1 Kd4 Ka1 Bf7 Kb1 Bb3 Ka1 Bf7 Kb1 Bb3 Ka1 Bf7 Kb1 Bb3 Ka1 
2.88 21/27 (-244cp) Kb2 Kg2 Ka3 Bb3 Kb2 Kf3 Ka3 Ke4 Kb2 Kd3 Ka1 a3 Kb1 Kd4 Ka1 Bf7 Kb1 Bd5 Ka1 Bb3 Kb1 Ke4 Ka1 Bc4 Kb1 Bd5 Ka1 
2.41 20/25 (-244cp) Kb2 Kg2 Ka3 Bb3 Kb2 Kf3 Ka3 Ke4 Kb2 Kd3 Ka1 a3 Kb1 Kd4 Ka1 Kd3 Kb1 Kd4 Ka1 Kd3 Kb1 Kd4 Ka1 Kd3 Kb1 
1.92 19/23 (-244cp) Kb2 Kg2 Ka3 Bb3 Kb2 Kf3 Ka3 Ke4 Kb2 Kd3 Ka1 a3 Kb1 Kd4 Ka1 Kd3 Kb1 Kd4 Ka1 Kd3 Kb1 Kd4 Ka1 
1.47 18/23 (-244cp) Kb2 Kg2 Ka3 Bb3 Kb2 Kf3 Ka3 Ke4 Kb2 Kd3 Ka1 a3 Kb1 Kd4 Ka1 Kd3 Kb1 Kd4 Ka1 Kd3 Kb1 Kd4 Ka1 
1.08 17/21 (-244cp) Kb2 Kg2 Ka3 Bb3 Kb2 Kf3 Ka3 Ke4 Kb2 Kd3 Ka1 a3 Kb1 Kd4 Ka1 Kd3 Kb1 Kd4 Ka1 Kd3 Kb1 
0.82 16/19 (-184cp) Kb2 Kg2 Ka3 Bb3 Kb2 Kf3 Ka3 Ke4 Kb2 Kd3 Ka1 Kc3 Kb1 Ba2 Kxa2 Kd4 Kb2 
0.64 15/18 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 Kg2 Ka1 Kf3 Kb1 Kg2 Ka1 Kf3 Kb1 Kg2 
0.55 14/17 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 Kg2 Ka1 Kf3 Kb1 Kg2 Ka1 Kf3 Kb1 
0.46 13/17 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 Kg2 Ka1 Kf3 Kb1 Kg2 Ka1 Kf3 Kb1 
0.41 12/15 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 Kg2 Ka1 Kf3 Kb1 Kg2 Ka1 
0.34 11/15 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 Kg2 Ka1 Kf3 Kb1 Kg2 Ka1 
0.27 10/15 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 Kg2 Ka1 Kf3 Kb1 Kg2 Ka1 
0.23 9/13 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 Kg2 Ka1 Kf3 Kb1 
0.19 8/10 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 Kb2 
0.14 7/9 (0cp) Kb2 Kg2 Ka3 Kf3 Kxa4 Kg2 Ka3 Kf3 
0.09 6/9 (0cp) Kb2 Kg2 Kb1 Kf3 Ka1 Kg2 Kb2 Kf3 Ka1 
0.07 5/9 (0cp) Kb2 Kg2 Kb1 Kf3 Ka1 
0.04 4/7 (0cp) Kb2 Kg2 Kb1 Kf3 
0.02 3/7 (0cp) Kb2 Kg2 Kb1 
0.02 2/3 (0cp) Kb2 Kg2 
0.01 1/1 (0cp) Kb2 

http://op12no2.me/toys/lozzadev/fen.htm ... 0-%200%201 <- NB: IF YOU CLICK ANALYSE IT USES RESOURCES ON YOUR MACHINE NOT A SERVER

Which makes me think search is broken...!

[Ferdy]

What positions do you recognize as draw in KQKP? This is one I haven't implemented yet.

I only have a/h pawn and c/f pawn. For other files, the Q usually wins and I don't implement them.

I have this book as a guide for kqkp bishop-pawns.
For defending king on right side, meaning the king is close to rook file, the offensive king must be 1 distance away from a specific squares in order to win.
In the diagram (black to move) black wins because its king is close to one of the squares marked with 1.




When the defending king is on the wrong side a distance of 2 squares from offensive king to specific squares marked 2, is enough to win the game. Black to move.



Pawns on rook files are also described in similar method of square numbering in the book.

Book ref.
100 Endgames You Must Know by Jesus de la Villa
Third Edition, New In Chess 2012.