Yep. There is even a Novotny move required to defuse the *spoiler* >selfstalemate attempt!<
Studies to Troll your engines with
Sort:
Apr 8, 2023
0
#43
Rocky64 wrote:
drdos7 wrote:
The first Position is one that I used a 5 mover from 1904 by Friedrich Köhnlein which you can see here:
The funny thing about the original 5 mover is that most engines don't opt for the mate in 5, and instead opt for the FREE ROOK!!!! which results in a mate in at most 19 instead.
That's a brilliant 5-mover! It took me about 20 minutes to solve it, including finding Black's best defence (which involves *spoiler* >a stalemate attempt<). A few years ago I wrote some blogs about composed problems that are short but couldn't be solved by Stockfish. These blogs are kind of dated because newer versions of SF managed to crack them. But the SF14 on Lichess does seem unable to solve this composition by Köhnlein. This is a great find since it's only 5 moves long and could be a record for the shortest forced-mate unsolved by SF. I also have SF14 on my PC and might download SF15 just to test it with this problem!
Hello Rocky, I'm afraid You'll get the same result with Stockfish 15
Apr 8, 2023
0
#44
Position #7
The engine troll factor on this one is about 8.5-9/10
White to play and DRAW
drdos7 wrote:
The first Position is one that I used a 5 mover from 1904 by Friedrich Köhnlein which you can see here:
I heavily modified it into one that I was trying to stump engines with which became Position #1, and I of course tested it for cooks and checked to make sure a few top engines couldn't solve it. The funny thing about the original 5 mover is that most engines don't opt for the mate in 5, and instead opt for the FREE ROOK!!!! which results in a mate in at most 19 instead.
Chess.com's SF15 NNUE which presumably runs in my browser solves this 5-mover in about 4 minutes. Quite decent I think. From the start it displays 1. fxe5 and 1.Kc1 as the main key move candidates.
When I saw the diagram for the first time, my first thought was: "Wouldn't it be nice to have a selfstalemate around square a4? Guess I'll have to make that". Didn't know yet that was the core line of the real solution and I started composing "my own version" before solving this one. Then I hit on Rocky's post which enlightened me. Once again I had landed in a synchronicity event with a peer composer. Happens a lot to me!
Btw, this type of selfstalemate is suitable for problems and much less for endgame studies. In an endgame study white can undo the selfstalemate cheaply by capturing bRa5 at any time. But in a direct mate problem the challenge is all about solution length and if the selfstalemate threat delays the mate by just one move, it is a resounding success. That's why it works here!
Apr 8, 2023
0
#47
Arisktotle wrote:
drdos7 wrote:
The first Position is one that I used a 5 mover from 1904 by Friedrich Köhnlein which you can see here:
I heavily modified it into one that I was trying to stump engines with which became Position #1, and I of course tested it for cooks and checked to make sure a few top engines couldn't solve it. The funny thing about the original 5 mover is that most engines don't opt for the mate in 5, and instead opt for the FREE ROOK!!!! which results in a mate in at most 19 instead.
Chess.com's SF15 NNUE which presumably runs in my browser solves this 5-mover in about 4 minutes. Quite decent I think. From the start it displays 1. fxe5 and 1.Kc1 as the main key move candidates.
You are correct here, but only if you use the default "pricipal variations"or "pv" parameter OR "LINES" as chess.com's version of Stockfish calls it in the engine =3 , here it is on my computer (meaning that I have a GUI installed LOCALLY on my computer with Stockfish 15 installed LOCALLY, rather than using the online version on chess.com) with "Lines"=3 or more correctly called pv=3.
22/35 00:45 1,319,242k 28,774k +17.65 1.a3-a4 Re5-e7 2.a4-a5 b6xa5 3.Nc7-d5 Re7-e6 4.b4xa5 Bd7-c6 5.Nd5-b6 Re6-e5 6.f4xe5 f5-f4 7.Ng2xf4 Kg4xh4 8.e5-e6 Kh4-g4 9.Nf4-d5 Kg4-f5 10.Nd5-e7+ Kf5xe6 11.Ne7xc6 Ke6-f6 12.Nc6-d4 Kf6-g6 13.Bb2-a3 h5-h4 14.Kb1-b2 Kg6-g5 15.Nb6xc4 Kg5-g4 16.Ba3-d6 Kg4-g5 17.Nc4-e5 Kg5-h5 18.Kb2xb3
22/37 00:45 1,319,242k 28,774k +23.11 1.f4xe5 Bd7-c6 2.e5-e6 f5-f4 3.Ng2xf4 Kg4xf4 4.e6-e7 Bc6-d7 5.e7-e8Q Bd7xe8 6.Nc7xe8 Kf4-e5 7.Kb1-c1 Ke5-d5 8.Ne8-f6+ Kd5-e6 9.Nf6xh5 Ke6-f5 10.Nh5-g3+ Kf5-f4 11.h4-h5 a6-a5 12.b4-b5 a5-a4 13.Re3xe4+ Kf4xg3 14.Re4xc4 Kg3-f3 15.h5-h6 Kf3-f2 16.h6-h7 Kf2-g1 17.Rc4xa4
22/38 00:45 1,319,242k 28,774k +24.41 1.Kb1-c1 a6-a5 2.f4xe5 a5xb4 3.e5-e6 Bd7-c6 4.c3xb4 f5-f4 5.b4-b5 f4-f3 6.b5xc6 f3xg2 7.Re3-e1 Kg4-g3 8.Bb2-d4 Kg3xh4 9.e6-e7 e4-e3 10.Bd4xe3 c4-c3 11.d2xc3 b3-b2+ 12.Kc1xb2 d3-d2 13.Re1-d1 g2-g1Q 14.Be3xg1 Kh4-g5
---------------------------------------------------------------------------
23/38 01:06 1,938,322k 29,202k +18.83 1.a3-a4 Bd7xa4 2.f4xe5 Ba4-c6 3.e5-e6 f5-f4 4.Ng2xf4 Kg4xh4 5.Nf4-g6+ Kh4-g5 6.Ng6-e5 Bc6-a4 7.Ne5xc4 h5-h4 8.Nc4xb6 Ba4-c6 9.c3-c4 Kg5-g4 10.Nb6-d7 Kg4-f5 11.Nd7-e5 Bc6-e8 12.Nc7xe8 Kf5xe6 13.Bb2-d4 Ke6-f5 14.Ne8-c7 Kf5-g5 15.Kb1-b2 Kg5-f6 16.Nc7xa6 Kf6-e7 17.b4-b5 h4-h3 18.b5-b6 h3-h2
23/46 01:06 1,938,322k 29,202k +26.30 1.f4xe5 Bd7-a4 2.e5-e6 Ba4-c6 3.Kb1-c1 f5-f4 4.Ng2xf4 Kg4xf4 5.e6-e7 Kf4-f5 6.e7-e8Q Bc6xe8 7.Nc7xe8 Kf5-e5 8.Kc1-b1 Ke5-d5 9.Ne8-f6+ Kd5-e5 10.Nf6-d7+ Ke5-d6 11.Nd7xb6 Kd6-c6 12.Nb6xc4 Kc6-d5 13.Nc4-a5 Kd5-e5 14.c3-c4+ Ke5-f5 15.c4-c5 Kf5-g4 16.Re3xe4+ Kg4-f3 17.Re4-c4 Kf3-g2 18.c5-c6 Kg2-h3 19.a3-a4 Kh3-g3 20.Na5xb3 Kg3-f2 21.Kb1-c1 Kf2-g2 22.Rc4-e4
23/40 01:06 1,938,322k 29,202k +26.30 1.Kb1-c1 Bd7-b5 2.f4xe5 Bb5-c6 3.e5-e6 f5-f4 4.Ng2xf4 Kg4xf4 5.e6-e7 Kf4-f5 6.e7-e8Q Bc6xe8 7.Nc7xe8 Kf5-e5 8.Kc1-b1 Ke5-d5 9.Ne8-f6+ Kd5-e5 10.Nf6-d7+ Ke5-d6 11.Nd7xb6 Kd6-c6 12.Nb6xc4 Kc6-d5 13.Nc4-a5 Kd5-e5 14.c3-c4+ Ke5-f5 15.c4-c5 Kf5-g4 16.Re3xe4+ Kg4-f3 17.Re4-c4 Kf3-g2 18.Kb1-c1 Kg2-f1 19.c5-c6 Kf1-f2 20.Na5xb3 Kf2-g1
---------------------------------------------------------------------------
23/38 02:28 4,243,244k 28,557k +18.83 1.a3-a4 Bd7xa4 2.f4xe5 Ba4-c6 3.e5-e6 f5-f4 4.Ng2xf4 Kg4xh4 5.Nf4-g6+ Kh4-g5 6.Ng6-e5 Bc6-a4 7.Ne5xc4 h5-h4 8.Nc4xb6 Ba4-c6 9.c3-c4 Kg5-g4 10.Nb6-d7 Kg4-f5 11.Nd7-e5 Bc6-e8 12.Nc7xe8 Kf5xe6 13.Bb2-d4 Ke6-f5 14.Ne8-c7 Kf5-g5 15.Kb1-b2 Kg5-f6 16.Nc7xa6 Kf6-e7 17.b4-b5 h4-h3 18.b5-b6 h3-h2
24/39 02:28 4,243,244k 28,557k +29.40 1.f4xe5 Bd7-a4 2.e5-e6 Ba4-c6 3.Kb1-c1 f5-f4 4.Ng2xf4 Kg4xf4 5.e6-e7 Kf4-f5 6.e7-e8Q Bc6xe8 7.Nc7xe8 Kf5-e5 8.Kc1-b1 Ke5-d5 9.Ne8-f6+ Kd5-e5 10.Nf6-d7+ Ke5-d6 11.Nd7xb6 Kd6-c6 12.Nb6xc4 Kc6-d5 13.Nc4-a5 Kd5-e5 14.c3-c4+ Ke5-f5 15.c4-c5 Kf5-f4 16.a3-a4 Kf4-f5 17.c5-c6 Kf5-g4 18.Re3xe4+ Kg4-f3 19.Re4-c4 Kf3-g2 20.c6-c7
24/46 02:28 4,243,244k 28,557k +M5 1.Kb1-c1 a6-a5 2.Kc1-d1 a5xb4 3.Kd1-e1 b4xa3 4.Ke1-f2 a3xb2 5.Re3-g3+
Apr 8, 2023
0
#48
Position #8
The engine troll factor on this one is 10/10
START this study AFTER balck's first move, otherwise 1.Rxe4 also wins, White to play and WIN AFTER 1.Nd5 Kxd5
Apr 8, 2023
0
#49
Arisktotle wrote:
Yep. There is even a Novotny move required to defuse the *spoiler* >selfstalemate attempt!<
Speaking of a Novotny, check this study out...it has one of the most beautiful Novotny moves I have ever seen. I call this study "The Immortal Novotny". It is not hard for engines, and I'm pretty sure you'll see it.
White to move and mate in 12
Apr 8, 2023
0
#50
Arisktotle wrote:
drdos7 wrote:
The first Position is one that I used a 5 mover from 1904 by Friedrich Köhnlein which you can see here:
I heavily modified it into one that I was trying to stump engines with which became Position #1, and I of course tested it for cooks and checked to make sure a few top engines couldn't solve it. The funny thing about the original 5 mover is that most engines don't opt for the mate in 5, and instead opt for the FREE ROOK!!!! which results in a mate in at most 19 instead.
Chess.com's SF15 NNUE which presumably runs in my browser solves this 5-mover in about 4 minutes. Quite decent I think. From the start it displays 1. fxe5 and 1.Kc1 as the main key move candidates.
That doesn't seem to be the case when I used Chess.com's SF15 NNUE on Chrome, so not sure what you mean when you said that the engine solved it in 4 minutes. Here's a screenshot of the SF15 evaluation after 1 hour at unlimited depth (my PC specs is pretty decent). The correct first move is the top pick but its eval is +26, not M5. The engine doesn't see White's threat and doesn't follow through with it, despite choosing a weak defence for Black (1...a5).
The one-hour limit is pretty arbitrary, but I don't think it can be claimed that SF15 solved this 5-mover. If you got a M5 eval after a few minutes, most likely you played some correct moves on the Analysis board in the same session, and the engine "remembers" them with caching to get the right result.
Apr 8, 2023
0
#51
drdos7 wrote:
Rocky64 wrote:
drdos7 wrote:
The first Position is one that I used a 5 mover from 1904 by Friedrich Köhnlein which you can see here:
The funny thing about the original 5 mover is that most engines don't opt for the mate in 5, and instead opt for the FREE ROOK!!!! which results in a mate in at most 19 instead.
That's a brilliant 5-mover! It took me about 20 minutes to solve it, including finding Black's best defence (which involves *spoiler* >a stalemate attempt<). A few years ago I wrote some blogs about composed problems that are short but couldn't be solved by Stockfish. These blogs are kind of dated because newer versions of SF managed to crack them. But the SF14 on Lichess does seem unable to solve this composition by Köhnlein. This is a great find since it's only 5 moves long and could be a record for the shortest forced-mate unsolved by SF. I also have SF14 on my PC and might download SF15 just to test it with this problem!
Hello Rocky, I'm afraid You'll get the same result with Stockfish 15
That's actually good news for me as I'm always looking out for relatively short problems that SF can't solve. I might write a blog about this 5-mover and if I do, I'll give you a mention for finding it. Actually, how did you find this composition as something that could be too hard for Stockfish?
Apr 8, 2023
0
#52
Rocky64 wrote:
drdos7 wrote:
Rocky64 wrote:
drdos7 wrote:
The first Position is one that I used a 5 mover from 1904 by Friedrich Köhnlein which you can see here:
The funny thing about the original 5 mover is that most engines don't opt for the mate in 5, and instead opt for the FREE ROOK!!!! which results in a mate in at most 19 instead.
That's a brilliant 5-mover! It took me about 20 minutes to solve it, including finding Black's best defence (which involves *spoiler* >a stalemate attempt<). A few years ago I wrote some blogs about composed problems that are short but couldn't be solved by Stockfish. These blogs are kind of dated because newer versions of SF managed to crack them. But the SF14 on Lichess does seem unable to solve this composition by Köhnlein. This is a great find since it's only 5 moves long and could be a record for the shortest forced-mate unsolved by SF. I also have SF14 on my PC and might download SF15 just to test it with this problem!
Hello Rocky, I'm afraid You'll get the same result with Stockfish 15
That's actually good news for me as I'm always looking out for relatively short problems that SF can't solve. I might write a blog about this 5-mover and if I do, I'll give you a mention for finding it. Actually, how did you find this composition as something that could be too hard for Stockfish?
I'm kind of an "engine specialist" if there were such a thing, and I have plenty of positions that are difficult for engines, I also know of some specialized UCI engines (they are mostly derivatives of Stockfish, but the code has been tweaked to make them much better at solving studies and mate finding). I've been around computer chess for over 30 years now and I actually know some of the programmers (no I haven't met any of them personally, but I have spoke with them on the forums and was actuually a beta tester for Stockfish very early in the game like from version 1.0) and I hang around the forums for computer chess, but to answer your question, I honestly "stumbled" upon that position searching for another position through google.
Here is a very nice Mate in 12 that Stockfish seems clueless on:
@Rocky64 and @drdos7: Though I work with the engines and tablebases I am not a technical engine specialist at all. As my hp-PC is 8 years old and its storage capacity down to a few Gigabytes, I refrain from installing new software on it. So I simply use the online engines offered by chess.com's analysis modules. No parameters and settings beyond the GUI ones like "unlimited time" - whatever that means. Which has the disadvantage of never seeing what happens exactly on the other side of the line. I was recently told that chess.com's SF engine is the browser version that completely runs on my local PC (don't know if that is true) . Which removes the unpredictabilities introduced by sharing resources on chess.com. What I noticed is that the browser version appears to keep intermediate results in cache memory, accessing them on multiple analysis jobs. Whatever is the case, I restarted my PC before getting the 4 minutes outcome which presumably clears the history data. Also I ran the analysis stand-alone by closing all other PC-applications to guarantee all computer resources are mine. Chess.com has been messing about with the engines in the past months. We had 2 SF15 versions + Komodo, then the 2 SF-versions were retracted and replaced by SF11, and recently the two SF15s have returned, SF15 NNUE and SF15 (faster). These changes were probably related to the overload problems on chess.com in jan/feb - due to some userbase merger? As a human solver I know that calculation results also have a "chance" component in the order of assessing candidate moves. If an engine decides to "roll the dice" on some occasions it may strike gold on one roll and not on another. Finally, the endpoints of analyses are undefined as SF doesn't know which goal to meet. My 5-mover analysis lasted 4 minutes because I watched the display lines and high fived when I saw "+M5". But meanwhile my engine continued its race through the galaxy in search of the next frontier "+M4"
. It's probably finished at some point but my analysis GUI never tells me
Repeat test (unfinished due to lack of time):
I repeated the 5-mover test today. SF15 NNUE this time was much slower which makes me suspect that I mouse-slipped on engine selection for the first one. Komodo solved it in 3 minutes repeatedly, SF15 NNUE in 10-30 minutes (couldn't track the analysis on the screen). Have to interrupt testing here due to other obligations. Btw, don't know if Komodo runs in the browser or on the server but I took the same precautions as for SF15 NNUE.
Apr 9, 2023
0
#56
Arisktotle wrote:
@Rocky64 and @drdos7: Though I work with the engines and tablebases I am not a technical engine specialist at all. As my hp-PC is 8 years old and its storage capacity down to a few Gigabytes, I refrain from installing new software on it. So I simply use the online engines offered by chess.com's analysis modules. No parameters and settings beyond the GUI ones like "unlimited time" - whatever that means. Which has the disadvantage of never seeing what happens exactly on the other side of the line. I was recently told that chess.com's SF engine is the browser version that completely runs on my local PC (don't know if that is true) . Which removes the unpredictabilities introduced by sharing resources on chess.com. What I noticed is that the browser version appears to keep intermediate results in cache memory, accessing them on multiple analysis jobs. Whatever is the case, I restarted my PC before getting the 4 minutes outcome which presumably clears the history data. Also I ran the analysis stand-alone by closing all other PC-applications to guarantee all computer resources are mine. Chess.com has been messing about with the engines in the past months. We had 2 SF15 versions + Komodo, then the 2 SF-versions were retracted and replaced by SF11, and recently the two SF15s have returned, SF15 NNUE and SF15 (faster). These changes were probably related to the overload problems on chess.com in jan/feb - due to some userbase merger? As a human solver I know that calculation results also have a "chance" component in the order of assessing candidate moves. If an engine decides to "roll the dice" on some occasions it may strike gold on one roll and not on another. Finally, the endpoints of analyses are undefined as SF doesn't know which goal to meet. My 5-mover analysis lasted 4 minutes because I watched the display lines and high fived when I saw "+M5". But meanwhile my engine continued its race through the galaxy in search of the next frontier "+M4". It's probably finished at some point but my analysis GUI never tells me
It is true that the engine versions on chess.com use local resources, however it's only capable of using 1 core or 1 processor at a time. I too have an 8 year old workstation, but it has 20 cores which Stockfish and other engines like Komodo can take advantage of by using all 20 cores at a time thereby searching much faster than the chess.com version. I can also control the size of the hash table, my computer has 32GB of RAM so I can use up to 16GB for the hash table size by using a local version of the engines. Another thing I should point out is that chess.com's default parameters make the engines weaker by searching 3 lines at time thereby splitting resources on already limited hardware (remember chess.com's engines only use 1 "core" or "thread". I should also point out that using a free GUI with Stockfish installed locally doesn't take up hardly any hard drive space... less than 100MB and probably less than 75MB. Using the local GUI also allows you to "copy and paste" the engine analysis like this:
FEN: nr1br3/4n1pK/1p2k1p1/P7/1Pb1N1N1/8/3P1Q2/8 w - - 0 1
Crystal 5 KWK avx2:
classical evaluation enabled
40/15 01:21 2,078,135k 25,530k 0.00 1.Ne4-g5+ Ke6-d5 2.Qf2-f3+ Kd5-d6 3.Qf3-f4+ Kd6-c6 4.Qf4xb8 Bc4-g8+ 5.Kh7xg7 Kc6-b5 6.Qb8-e5+ Kb5-a6 7.Qe5-b8
in that example we can see that I used the engine Crystal 5 KWK avx2 with the classical evaluation enabled meaning that I disabled the NNUE (or Neural Network file) and that it searched over 2 billion positions in 1 minute and 21 seconds at an average speed of over 25 million positions a second, and that it still hasn't solved the mate in 12 in the FEN
And how does that explain that chess.coms engines solve the 5-mover relatively fast on my PC?
Apr 9, 2023
0
#58
Arisktotle wrote:
And how does that explain that chess.coms engines solve the 5-mover relatively fast on my PC?
It doesn't, but if you had a local version installed it would solve it faster because you could use more than 1 core. The only reason it "solved" the mate in 5 is because it was searching 3 lines and happened to stumble upon the solution in one of the lines. If you had the local version you could run up to 24 lines at the same time.
I love this topic, and I know a few if these:
If it's black to move here, stockfish cannot figure it that to draw, all you have to do is check with the rook, trade rooks, and then play bh3! This forces either the doubling or loss of the g pawn resulting in a drawn endgame due to the wrong colored bishop and doubled h pawns.
The first Position is one that I used a 5 mover from 1904 by Friedrich Köhnlein which you can see here:
The funny thing about the original 5 mover is that most engines don't opt for the mate in 5, and instead opt for the FREE ROOK!!!! which results in a mate in at most 19 instead.
That's a brilliant 5-mover! It took me about 20 minutes to solve it, including finding Black's best defence (which involves *spoiler* >a stalemate attempt<). A few years ago I wrote some blogs about composed problems that are short but couldn't be solved by Stockfish. These blogs are kind of dated because newer versions of SF managed to crack them. But the SF14 on Lichess does seem unable to solve this composition by Köhnlein. This is a great find since it's only 5 moves long and could be a record for the shortest forced-mate unsolved by SF. I also have SF14 on my PC and might download SF15 just to test it with this problem!