Why 4...Kg4? Why not 4... Nf7+, fork the queen?
The puzzle that you, Carlsen nor Stockfish can solve
Sort:
Komodo 8 says White is better with no help. Yet another reason Stockfish is inferior.
1. Nf6+Kg72. Nh5+Kg63. Bc2+Kxh5Now we promote to a queen
4. d8=QIf black choose to take the queen by playing Nf7+, it's going to be mate in 11 for white, so best move for black is Kg4
4... Kg45. Qf6Kxg36. Qe5+Kf37. Bd1+Kg28. Qxe3c4+9. Kd5Bc510. Qxc3Bf211. Qf6Nc512. Qxh6Nh313. Kxc4Ng114. Qc6+Kh215. Bh5Be316. Qd5Kh317. Qf5+Kg218. Be8Kg319. Bc6Na620. Qg6+Kf421. Qe4+Kg522. Qxe3+Kg623. Qxg1+Kf624. Qa1+Kf525. Qxa6Kf626. Bf3+Ke527. Qh6Kf528. Qg7Kf429. Bh5Ke430. Qf6Ke331. Qd4#This is what my computer with Stockfish 6 figured out. It can however probably be done more accurately with a stronger computer and longer calculation time.
Komodo 8 instantly
Firenzina 2.43 depth = 26
Equinox 3.2 depth = 28
Remember that next time someone tries to say Stockfish is the best chess engine.
My calculations is done one a small laptop though, I'm not sure how Stockfish would handle this puzzle on a stronger computer.
Lincolm wrote:
Why 4...Kg4? Why not 4... Nf7+, fork the queen?
Press the questionmark to see my notes
I let SF6 run to depth = 37 and it couldn't find it.
Stockfish is great because it's strong and it's free. I don't know what the proprietary engines cost, but if you're a stundent it's nice to know you can download a very strong engine without having to pay for it.
I downloaded a modified version of Stockfish called Sting, and it is made to solve puzzles. It gave a score of 4+ very quickly.
How do you know Carlsen cannot solve this puzzle ? 
I figured out the first few moves, though the rest is tough. And yes, I wonder how you know that Carlsen didn't solve this.
Of course he could 
I heard that Mikhail Tal took a walk after he was presented with this puzzle and came back with the solution (Kasparov and Karpov couldn't solve it.) Does anyone know if this is true or not?
>"Yet another reason Stockfish is inferior"
stockfish6 against komodo8: +11−3=32
http://www.computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?print=Details&each_game=1&eng=Stockfish%206%2064-bit%204CPU#Stockfish_6_64-bit_4CPU
KStalsberg wrote:
I heard that Mikhail Tal took a walk after he was presented with this puzzle and came back with the solution (Kasparov and Karpov couldn't solve it.) Does anyone know if this is true or not?
https://gameknot.com/room-fm.pl?home=2&th=1090
Komodo 8 see Nf6 after 4-5 seconds but Stockfish 6 doesn't see it at all moreover after Kg6 Stockfish 6 still evaluate position over -4.50 only after Bc2 there is change from over -4.50 to ... over +4.50 
It's completely absurd to give this Kg4 line, all the difficulty and beauty of the problem is to find the mating net.
Yeah I've seen this one: https://www.youtube.com/watch?v=ncdsvKP0pPA
I have Stockfish 6 and the evaluation changes from -4 to +3 after Bc2+.
Wow, talk about necromancy! But I will say a few generic things. Talk around this puzzle often annoys me because people spread lots of falsehoods about engines and the way engines work as soon as this puzzle comes up. I want to correct some misconceptions about engines and shed some light on why engines don't solve this puzzle.
First of all, engines have no problem looking deep, as many people assume when they see engines have trouble with this. Actually, even on my PC, stockfish easily sees 55 moves deep or more well beyond the checkmate but still gives black as winning. How is this possible?
Well if you do the math, you see that with an average of 30 legal moves per ply, 900 moves per turn, that means 400^55 turns is... 10^143 possible lines to calculate. Now wait a minute, that's not right. Even the fastest computer ever couldn't calculate that many variations before the heat death of the universe. That's because engines don't calculate that many lines. They uses a process called pruning to choose which lines to calculate and which ones to skip. Modern engines beat older engines not because they are faster, but because the software has improved- computers have gotten very very good at pruning. In today's world, the top engines are set apart by what lines they choose to search first rather than how fast they can search them.
So when my stockfish says it has searched 55 moves deep, what it really means is "I have isolated one line of play that seems best for white, and calculated that out to 55 moves." So in reality, you could leave stockfish on for eternity. It would calculate out to 100 moves, 200 moves, 400 moves and always say black is losing because it already missed the winning move. This is something special about stockfish, and the thing that made it #1 in the world and now #2 behind Komodo (but no doubt edging for #1). Stockfish is a strong engine now because its software prunes agressively and effectively. If you compare stockfish with another engine with the same position on the same machine, you will almost always find that stockfish digs much faster than the other chess engine.
This is amazing for chess games and blitz computer chess, and makes it the #1/#2 chess engine.
But what it is not good for is "engine crusher" puzzles. The thing that determines whether an engine can solve this puzzle is not how deep it can look or how fast it can think, but whether or not it prunes those critical winning moves or chooses to investigate them. Stockfish prunes so heavily that it is likely to skip over winning moves in puzzles like this one, other engines are not as likely to.
Depending on your interface, you can also set your engine's pruning manually. What I mean is that after stockfish looks 30 moves deep, I can tell my stockfish "Okay, rather than spending your processing power to look even deeper into the position, go back over lines you pruned before." This will make it see the winning move, even though with default settings on a default computer it may not.
Weaker engines, such as Rybka or Houdini, might see the line as well if they don't prune as heavily. It's all a matter of pruning.
So yes, engines can solve this puzzle, and no, stockfish is not weak or inferior because it could not solve it. It's simply a matter of the default settings. Engines can definitely see deeper than humans, and our only hope is to be more "creative" by investigating moves they skipped over.
Brought to you by the committee of explanations that are wrong in every specific way, but right in general
White to move, mate in X:
There is this somewhat well known puzzle that I keep seeing now and then. This puzzle is very hard, and unless you've seen it before, you most likely wont be able to solve it. With perfect play it's a mate in 30 or less for white. But if you give it to most players or even strong chess engines like Stockfish, they will most likely not be able to understand that white is better here. But it is white to move, and white wins.
I gave this problem to Stockfish 6 on my computer and it gave it as black is winning, when in fact it should be showing the oppsite, which it understands only after I help it with a few moves. Maybe if we discover the quantum computer soon, Stockfish will be able to see deep enough to calculate the mate. But for now, this remains one of the hardest puzzles out there.