Studys where engines wont help

Stockfish 6 solved the position shown in post #5 when I gave it 20 seconds of time to think. Powered by i7-4770. You guys may have to specify what you mean by "can't" ;) I didn't bother studying and testing the other positions yet but I'm confident that SF will succeed at them.

Modern computers are not powerful enough to allow chess programs to look through every single possible combination up to the next 5+ moves (takes insane amount of time). That is why they have to cut both depth and lines out of their search. The less time engine is given, the more it has to cut. The more it has to cut, the more likely it will make mistakes. Same with human players: you might see GM missing a mate in 1 in bullet.

The position in #5 is tricky for the computer because it requires it to see the fifty-move-rule up ahead. Human can just stare at the pattern and realize that it will be a draw without thinking through the exact 50 moves of how it's going to happen. Engines do utilize positional analysis but apparently they lack the solution for that type of case so they really have to crank up the search/evaluation time in order to spot the draw by raw calculation.

That is right, #11 Stockfish here also finds the solution in around 50 seconds. However, even after finding the right moves it gives -10 for black instead of 0.00 so it finds the right moves but it evaluates the position wrong.

It will probably have some problems finding Nc8+ in my Study as there are more pieces on the board.

But even if it does, without giving 0.00 it is not really helpful for practical analysis.

Any more studies anyone? Please contribute!

OK, this one is amazing since my stockfish couldn't find the winning move and continuation even after 35+ plies!!! It thinks black is better.

This is the position, white to move and win:
http://en.lichess.org/analysis/8/3P3k/n2K3p/2p3n1/1b4N1/2p1p1P1/8/3B4_w_-_-

This is what stockfish thinks after all that pondering:

Depth: 36
Evaluation: -1.82
Best move: 1.d8=Q
Main line: 1.d8=Q Nf7+ 2.Kd7 Nxd8 3.Kxd8 Ba5+ 4.Kd7 Bc7 5.Nxe3 Bxg3 6.Nd5 Be5 7.Kc6 Nb4+ 8.Nxb4 cxb4 9.Kc5 h5 10.Kxb4 Kg6 11.Ba4 h4 12.Kc4 Kg5 13.Kd3 h3 14.Bc6 Kf4 15.Ke2 Bf6 16.Bb7 Kg3 17.Be4 Be5 18.Ke3 Kh2 19.Kf2 Bd4+ 20.Kf1 Kg3 21.Ke2 Kf4 22.Bb7 Be5 23.Kd1

The solution is below, one of the variants at least which stockfish played here against me. It is actually close to the main continuation of the puzzle which is an artistically crafted composition. It forces black to underpromote to a knight a couple of times to prevent the checkmate.

http://en.lichess.org/8B7Japk7ZVBh

There is actually a better way to continue for black after queen promotion(4. ...Kg4) which my stockfish preferred to play but white still could clear the board with B&Q pair and easily won with accurate play.

#11

I haven't really tested this position with SF only with the latest version of fritz. I am pretty sure tho that an earlier version of SF wouldn't solve it. Maybe they added something or similiar games or a tablebase or something. Either way it's very nice to see that chess engines are progressing :)

#13 This is an amazing Puzzle. Thank you.

Hi.

Well, I do not know any middle game positions like that, however from my correspondence experience: weaker engines are extremely bad at closed positions there one side has a space adventage (for example king's indian). though best engines seem to be good at it.

Every engine, no matter how stong it is, does not understand positions with some kinds of doubled and isolated pawns (it helped me win a tournament with score of 7/7 against all players being engine users: I chose lines with doubled pawns and used them, spending time analysing myself, rather then believing engine). Fun thing is that engine's are also bad at some gambit's that even they can not calculate fully, because they seem to want retreat instead of continously attacking.

So openings that engine's are bad at, in my opinion are:

-King's Indian
-Ruy Lopez (especially berlin)
-Benoni
-Traxler counter attack (with B:f7)

Basically all (not buggy) engines will find all solutions to all Studies if you let it run enough longer. May be it will take depth 150 (or 300) to see it's draw by 50-moves rules after three (or 6) captures.
Saying an engines doesn't find a move is a big lie. Only the time to find (or not find) a solution is relevant.
If you want to see my latest set of "problematic" studies for engines, it's here : http://www.talkchess.com/forum/viewtopic.php?t=57056

Please everyone, my Post title was: "Studies where Engines wont help". And indeed: Engines are not helpful with all Studies uploaded here so far. They dont find the moves faster than a human, or if they do, they evaluate the position wrong.

I never said that engines can not find the solution. The solution can always be found. Even by entering random moves.

Position to blow up poor engine.

Created by Serhiy Didukh, Ukraine.

White to move and make a draw:

r4B1k/3p2Rp/7K/2pp4/1q3pBp/1N4b1/4R3/4r3 w - -

Solution:
(do not look yet, try to play yourself first)

http://en.lichess.org/04B8SGPsxBIT

Not a study, but perhaps the most famous endgame position ever:
http://en.lichess.org/wfVtDYyw#93
Stockfish gives _the only winning move_ as a mistake.

There's hope for humanity!