Forums

This puzzle is historically wrong!

Sort:
mohshenas

This controversial and strange-looking puzzle was created around 1970 by a Dutch game composer, Gijs Van Breukelen, and presented in a super tournament in which several grandmasters, notably Garry Kasparov and Mikhail Tal, had taken part. No one could solve the puzzle on the first try, but Tal did (?!). This puzzle is also known as Plaskett's Puzzle, named after Harold James Plaskett, a British chess grandmaster.

 

 

 

 

The strange thing about this puzzle is that the Black is the winner based on modern engines with -3.87 calculated by the Stockfish Chess Engine on depth 53 and -3.5 calculated by Stockfish on depth 54 (Cloud). Surprisingly, Tal's solution is one of the ways that win the game for the White. According to Tal's solution, the evaluation number of the Stockfish for the second move is reversed from -3 to +6! The local traps trick the Engines! Or can it be granted as a puzzle? Maybe it's just a position from a middle game.

Finally, I analyzed the position via a combined search technique; i.e., I combined the Stockfish and Lc0 in a way that the Lc0 was iteratively pruning the unfathomable bounds created by the Stockfish's Alpha-beta tree. So, the unpromising bounds were fathomed quickly. So, I calculated the first five optimal moves w.r.t. the Stockfish's Evaluation Function:

1. Nxe3 Ba5

2. Kc6 Nb8+

3. Kb5 Bd8

4. Bg4 Kg7

5. Kxc5 Bf6

And the combined method persistently says that the Black is winning!

BishopTakesH7
mohshenas wrote:

 

So, I calculated the first five optimal moves w.r.t. the Stockfish's Evaluation Function:

 

1. Nxe3 Ba5

2. Kc6 Nb8+

3. Kb5 Bd8

4. Bg4 Kg7

5. Kxc5 Bf6

And the combined method persistently says that the Black is winning!

 

I fail to understand what you mean. Nxe3?? Is a losing blunder. The solution is Nf6+.

BishopTakesH7

Of course, there is this blemish in the study:

 

 

 

But if I remember correctly, moving the black knight to h8 fixes the problem.

 

However, this isn't what you are stating is "wrong" with the problem, so what do you mean?

mohshenas
BishopTakesH7 wrote:

Of course, there is this blemish in the study:

 

 

 

But if I remember correctly, moving the black knight to h8 fixes the problem.

 

However, this isn't what you are stating is "wrong" with the problem, so what do you mean?

I searched the complete depth. This solution is not correct.

BishopTakesH7
mohshenas wrote:

I searched the complete depth. This solution is not correct.

 

What do you mean? This is the solution:

 

 

 

BishopTakesH7

Or the fixed solution, which even the engine can see:

 

 

BishopTakesH7

Scratch that, it can't see it.

Rocky64

All the OP has shown is that even modern engines have trouble cracking this famous study, which is in fact sound.

BishopTakesH7 wrote:

Of course, there is this blemish in the study:

1. Nf6+ Kg7 2. Nh5+ Kg6 3. Bc2+ Kxh5 4. d8=Q Kg4! {No mate!}

Actually, someone recently posted a Stockfish+tablebases line showing that even in the original setting, White does have a forced mate in 34 (at most) in the 4...Kg4 variation (with 5.Qf6). I can't verify that, but testing the position with Stockfish 14 (without TBs), after an hour it evaluates 5.Kc6 as +5.5 and 5.Qf6 as +5.08. Here's the 5.Kc6 variation that shows how White wins a minor piece soon after, to reach a tablebase position that's a definite win. 

5.Kc6 Nf3 6.Qg8+ Kh3 7.Qe6+ Kg2 8.Qxe3 Nd4+ 9.Kb7 Nxc2 10.Qe2+ Kxg3 11.Qxc2 Nb8 12.Kxb8.

 

 

yetanotheraoc

Die Schwalbe gives a win in the 5.Qf6 variation, although for some reason they have the typo 4...Kh4 (illegal move) instead of 4...Kg4.

https://pdb.dieschwalbe.de/search.jsp?expression=PROBID=%27P1379127%27

4...Kg4 is not really a flaw. The original task was not "mate" it was "win". It's pretty much a given if black lets the newly promoted queen live, white will win somehow.

mohshenas
yetanotheraoc wrote:

Die Schwalbe gives a win in the 5.Qf6 variation, although for some reason they have the typo 4...Kh4 (illegal move) instead of 4...Kg4.

https://pdb.dieschwalbe.de/search.jsp?expression=PROBID=%27P1379127%27

4...Kg4 is not really a flaw. The original task was not "mate" it was "win". It's pretty much a given if black lets the newly promoted queen live, white will win somehow.

The calculations are absolutely wrong! Please read my article!

Arisktotle
mohshenas wrote:

The calculations are absolutely wrong! Please read my article!

I'd say you are out of your depth on this one. None of the game playing engines (including lc0 and stockfish) is capable of beating a tablebase on its home soil since the latter produces only perfection. Also, reading depth does not imply a complete tree walk of all the variations. It's more like sinking a bathyscaphe to 30,000 feet under water. Proving it has has been there is not the same as stating it has seen everything there is to see on that depth or underway!

yetanotheraoc
mohshenas wrote:
yetanotheraoc wrote:

Die Schwalbe gives a win in the 5.Qf6 variation, although for some reason they have the typo 4...Kh4 (illegal move) instead of 4...Kg4.

https://pdb.dieschwalbe.de/search.jsp?expression=PROBID=%27P1379127%27

4...Kg4 is not really a flaw. The original task was not "mate" it was "win". It's pretty much a given if black lets the newly promoted queen live, white will win somehow.

The calculations are absolutely wrong! Please read my article!

I did read your "article" (post #1). Please read post #2 by BishopTakesH7, who shows why _your_ calculations are absolutely wrong, based on a wrong first move. The author, van Breukelen, claims a win for white after 1.Nf6+. If you want to claim a win for black, you need to provide black moves after 1.Nf6+ that improve on the variation given by van Breukelen. Showing black wins after 1.Nxe3 doesn't accomplish that.

mohshenas
yetanotheraoc wrote:
mohshenas wrote:
yetanotheraoc wrote:

Die Schwalbe gives a win in the 5.Qf6 variation, although for some reason they have the typo 4...Kh4 (illegal move) instead of 4...Kg4.

https://pdb.dieschwalbe.de/search.jsp?expression=PROBID=%27P1379127%27

4...Kg4 is not really a flaw. The original task was not "mate" it was "win". It's pretty much a given if black lets the newly promoted queen live, white will win somehow.

The calculations are absolutely wrong! Please read my article!

I did read your "article" (post #1). Please read post #2 by BishopTakesH7, who shows why _your_ calculations are absolutely wrong, based on a wrong first move. The author, van Breukelen, claims a win for white after 1.Nf6+. If you want to claim a win for black, you need to provide black moves after 1.Nf6+ that improve on the variation given by van Breukelen. Showing black wins after 1.Nxe3 doesn't accomplish that.

It is now proved to me that you barely have the knowledge of chess computations! A chess engine can suggest a move that no one is able to justify that! Please note that I have a Ph.D. in computer science and I have studied chess engines for several years.

mohshenas
Arisktotle wrote:
mohshenas wrote:

The calculations are absolutely wrong! Please read my article!

I'd say you are out of your depth on this one. None of the game playing engines (including lc0 and stockfish) is capable of beating a tablebase on its home soil since the latter produces only perfection. Also, reading depth does not imply a complete tree walk of all the variations. It's more like sinking a bathyscaphe to 30,000 feet under water. Proving it has has been there is not the same as stating it has seen everything there is to see on that depth or underway!

Please study the Monte-Carlo method. We do not need to go to the depth deterministically.

yetanotheraoc

Kindly provide your real name and what university granted you a Ph.D. in computer science. Then we will know which degree program and graduates thereof to avoid in the future.

tlay80

Anyway, a degree in reading might come in handier.

Graywing13

@mohshenas you seem to think while engines can suggest moves humans can't justify, humans can't suggest moves engines can't justify wink.png

How about this: you, using all available resources, play a game as black starting from that position. I'll play white.

If you accept, I'll start with Nf6+

porkqupine

So many words and waving your credentials around, when everything that's required of you to prove the puzzle wrong is to provide a line that refutes the 1.Nf6+ line (post #5 here). Specifically, which of black's moves in the solution is wrong and what is the right move then. You haven't done any of that yet. Might it be because you don't have the line required?

That was a rhetorical question.

mohshenas
yetanotheraoc wrote:

Kindly provide your real name and what university granted you a Ph.D. in computer science. Then we will know which degree program and graduates thereof to avoid in the future.

Go to my profile and google my full name.

mohshenas
Graywing13 wrote:

@mohshenas, you seem to think while engines can suggest moves humans can't justify, humans can't suggest moves engines can't justify

How about this: you, using all available resources, play a game as black starting from that position. I'll play white.

If you accept, I'll start with Nf6+

 

I openly accept your challenge!

What is the precise definition of the best move? Let me define it. The best move is the globally optimal path w.r.t. an Evaluation Function or a Fitness Function. Every chess engine operates via an evaluation function for calculating positions. For example, you can find the evaluation function for the Stockfish here.

I also evaluated the position via the Lc0 and the Stockfish in the chessify.me computing cluster, and the result is the same as mine! The first optimal move is again Nxe3.