mmt Vs. Ovyron (G4 D5 BG2)

[jp]

[d]4b1k1/8/5n1r/4N3/8/5Q2/3K4/8 w - - 0 1

There is only one winning move with DTZ<100 (There are some moves that are practically a draw by the 50 move rule because DTZ>=100 that is called a cursed win)
<snip>
Maybe unassisted engine without tablebases can find the right line and I did not test it.

For the closely related position

6N1/3n4/3k1b2/8/1r6/5K1Q/8/8 w - - 0 1 ,

Stockfish (with 6-man tables) has Qf5 with eval +0.38 at depth 54. Wrong move, and eval is off by infinity!

but

Cfish has done better. It has the right move, though its current evaluation is not at all convincing. Of course, Cfish is currently using 6-man tables. I should disable those.

+0.72 1. Kg2 Rb2+ 2. Kf1 Rb1+ 3. Ke2 Rb2+ 4. Kd1 Rb1+ 5. Kc2 Rb2+ 6. Kc1 Rb5 7. Qh2+ Re5 8. Qh6 Rc5+ 9. Kd2 Rd5+ 10. Ke2 Re5+ 11. Kd3 Rd5+ 12. Ke4 Nc5+ 13. Ke3 Re5+ 14. Kd2 Ne4+ 15. Kd3 Re6 16. Qf4+ Kc6 17. Qf3 Bg5 18. Kd4 Bd8 19. Qf5 Ng5 20. Kc4 Re4+ 21. Kc3 Re6 22. Kd2 Ne4+ 23. Kc2 Nc5 24. Qh5 Kb5 25. Kd1 Ba5 26. Nh6 Re1+ 27. Kc2 Bc7 28. Nf5 Re4 29. Kd1 Kb4 30. Qf7 Be5 31. Qf8 Bf4 32. Qa8 Kb5 33. Qd5 Be5 34. Kd2 Kb4 35. Kc2 (depth 54, 0:30:43)

(Yes, TBs should not be used.)

[zullil]

Of course it is not mate in 3 so 3.Rxf5 is not mate but the idea is simply to post a single line when white may have faster mate but if black does not follow the single line you then white can also mate in at most 35 moves.

But I didn't save the line, and it has been overwritten to a mate in 34 already. Some other mate in 35 line wouldn't work because Zullil wants to see the one that I found in 1.5 days, this mate in 34 was found after 17 days. And, anyway, I'm checking these lines for correctness myself, their correctness is irrelevant because the only correctness that matters is from the line I'll send for the challenge. Zullil kind of spoiled the challenge after announcing a mate in 28 (so I know what not to send), but I have not found a line to send anyway (all the lines I've seen could be improved trivially. Who knows if I'll make it to 28.)

To prove 1.Rh8+ is a mate in 28 you only need to solve 2 positions to mate in at most 26 moves

1.Rh8+ Kf7 2.Rf1+ Qf6
1.Rh8+ Kf7 2.Rf1+ Qf5

I believe that I have a solution for the first one that has to be a correct line by my definition assuming there is a solution for the second line because black cannot delay the mate to more than 28 moves by choosing the second line.

This whole thing has become rather ridiculous. Why Ovyron is even continuing is beyond me. At the pace he's going, most of us will be dead before he even reaches mate in 30. Much less mate in 28. Which, by the way, is not minimal. At least based on a private message from someone I trust ...

17 days to get to mate-in-34 is beyond terrible. This would take Stockfish maybe eight hours on a strong system, probably with no human guidance at all.

I was way off. About 45 minutes to get a mate-in-34, with no guidance and a full PV (not shown):

info depth 81 seldepth 68 multipv 1 score mate 34 nodes 234620749765 nps 86244184 hashfull 999 tbhits 0 time 2720424 pv h3h8 g8f7 h1f1 [SNIP]

And first mate score in less than 20 minutes:

info depth 61 seldepth 74 multipv 1 score mate 51 lowerbound nodes 93486215208 nps 82113856 hashfull 970 tbhits 0 time 1138495 pv h3h8

[Zenmastur]

If it takes you days to find an optimal line of play in this setting, things can only be worse in "typical" positions

Nope, I found the optimal line within minutes.

Do you think you found the optimal line of play when playing Harvey?

I'll answer the question for you. You didn't. You happened to run into a transposition that let you get into a good line of play. If it weren't for this fortuitous transposition your position could have been much worse. It's unlikely that you could see this in a “reasonable” period of time with your hardware.

The point is you don't know what you don't know! I do know because I've played on hardware similar in performance to yours. That's why I've upgraded my computers several times times since I started playing chess again.

To find a best line of play, one always needs to consider all possible defenses by the opponent.

I checked all defenses of the opponent in 1.5 days, leading to a mate in 35 that black couldn't delay further, in a normal game that would have been enough.

The difference between this test and a “normal” position from a game is that it's intuitively obvious when the goal has or hasn't been reached. In an opening or a middle game where the position is about equal you have no way to know if you have reached your objective before making a move. You're just hoping/guessing that you did enough analysis. The engine's evaluation will give you little information about the “true” value of the position unlike a mate in X problem. This is why you played the way you did in your game against Harvey. You couldn't see deep enough to “know” the line you ended up playing wasn't going to get you where you wanted to go.

BTW, I could afford better hardware, sure. I have to because I don't know how long my CPU will last, it's getting 10 years old this year, and there's no way I'm going back to my single CPU 32bit. But I'll make the jump because I'll have no other alternative, not because I want to. And if it lasts another 10 years, well, there's things I don't have yet that I'm considering buying instead (such as... giants speakers? Yeah, my neighbors sometimes are very noisy with theirs, though apparently they rent them for parties. Something I can't do with my hardware is listening to music at that level, or listening to movies at the level of cinema's speakers, so it'd make sense to upgrade those.)

Really? You would buy speakers to annoy your neighbors but you won't upgrade your system? YOU'RE REALLY SICK!

Do you spend more time on your computer or more time annoying your neighbors? Because curious minds want to know!

But millions of people (who may well want to play CC) cannot. This general question is surely more interesting than, for example, some billionaire's ability to buy a supercomputer.

My claim is that this billionaire can save some bucks by learning better analysis methods. Methods that would allow them to find the only winning line in a position to beat their opponent. Methods that would allow them to find the only saving move in a position they were losing. They don't need better hardware for this and thus buying it would be a waste of money, regardless of how much they have.

Some people don't have huge amounts of time to waste! Which is exactly what you're doing when using slow hardware! Time is money. If you earn $20 an hour and you waste 300 hours on a problem that would take 10 hours on a “cheap” but modern system you have wasted 290 hours of your time. 290 hours at $20 per hour is $5800. You could spend the 290 hours you wasted earning more money to buy a system OR you could just spend all that extra time playing more games/ solving more problems etc. The point is you would have “at least” 290 hours of time to spend as you like.

Who do you think is smarter the billionaire that “wastes” his money on a nice computer system or you wasting hundreds of hours on a problem that everyone else can solve in an evening?

Imagine what you could do with thousands of hours of “free” time a newer computer would grant you over it's useful life!

A perfect example:

I was way off. About 45 minutes to get a mate-in-34, with no guidance and a full PV (not shown):

info depth 81 seldepth 68 multipv 1 score mate 34 nodes 234620749765 nps 86244184 hashfull 999 tbhits 0 time 2720424 pv h3h8 g8f7 h1f1 [SNIP]

And first mate score in less than 20 minutes:

info depth 61 seldepth 74 multipv 1 score mate 51 lowerbound nodes 93486215208 nps 82113856 hashfull 970 tbhits 0 time 1138495 pv h3h8

Regards,

Zenmastur

[mmt]

I found a couple of test matches of the LC0 weights net I'm using and I wrote a program to extract evaluation data and positions from pgns. LC0 is better than SF 11 when the sum of "firepower" on the board (1 point for pawn, 3 for bishop/knight, 5 for rook, 9 for queen) is around 60 or more https://i.ibb.co/zN4bn2d/ff.png. It's worse with less firepower. This game is at 64 right now so it's still worth it to run LC0.

If you wanted to create a hybrid program, I think you could find a more sophisticated measure. I think a hybrid that chooses what lines to explore would be better still. I think there is probably a room for 30+ Elo improvement.

[zullil]

If it takes you days to find an optimal line of play in this setting, things can only be worse in "typical" positions

Nope, I found the optimal line within minutes.

Do you think you found the optimal line of play when playing Harvey?

I'll answer the question for you. You didn't. You happened to run into a transposition that let you get into a good line of play. If it weren't for this fortuitous transposition your position could have been much worse. It's unlikely that you could see this in a “reasonable” period of time with your hardware.

The point is you don't know what you don't know! I do know because I've played on hardware similar in performance to yours. That's why I've upgraded my computers several times times since I started playing chess again.

To find a best line of play, one always needs to consider all possible defenses by the opponent.

I checked all defenses of the opponent in 1.5 days, leading to a mate in 35 that black couldn't delay further, in a normal game that would have been enough.

The difference between this test and a “normal” position from a game is that it's intuitively obvious when the goal has or hasn't been reached. In an opening or a middle game where the position is about equal you have no way to know if you have reached your objective before making a move. You're just hoping/guessing that you did enough analysis. The engine's evaluation will give you little information about the “true” value of the position unlike a mate in X problem. This is why you played the way you did in your game against Harvey. You couldn't see deep enough to “know” the line you ended up playing wasn't going to get you where you wanted to go.

BTW, I could afford better hardware, sure. I have to because I don't know how long my CPU will last, it's getting 10 years old this year, and there's no way I'm going back to my single CPU 32bit. But I'll make the jump because I'll have no other alternative, not because I want to. And if it lasts another 10 years, well, there's things I don't have yet that I'm considering buying instead (such as... giants speakers? Yeah, my neighbors sometimes are very noisy with theirs, though apparently they rent them for parties. Something I can't do with my hardware is listening to music at that level, or listening to movies at the level of cinema's speakers, so it'd make sense to upgrade those.)

Really? You would buy speakers to annoy your neighbors but you won't upgrade your system? YOU'RE REALLY SICK!

Do you spend more time on your computer or more time annoying your neighbors? Because curious minds want to know!

But millions of people (who may well want to play CC) cannot. This general question is surely more interesting than, for example, some billionaire's ability to buy a supercomputer.

My claim is that this billionaire can save some bucks by learning better analysis methods. Methods that would allow them to find the only winning line in a position to beat their opponent. Methods that would allow them to find the only saving move in a position they were losing. They don't need better hardware for this and thus buying it would be a waste of money, regardless of how much they have.

Some people don't have huge amounts of time to waste! Which is exactly what you're doing when using slow hardware! Time is money. If you earn $20 an hour and you waste 300 hours on a problem that would take 10 hours on a “cheap” but modern system you have wasted 290 hours of your time. 290 hours at $20 per hour is $5800. You could spend the 290 hours you wasted earning more money to buy a system OR you could just spend all that extra time playing more games/ solving more problems etc. The point is you would have “at least” 290 hours of time to spend as you like.

Who do you think is smarter the billionaire that “wastes” his money on a nice computer system or you wasting hundreds of hours on a problem that everyone else can solve in an evening?

Imagine what you could do with thousands of hours of “free” time a newer computer would grant you over it's useful life!

A perfect example:

I was way off. About 45 minutes to get a mate-in-34, with no guidance and a full PV (not shown):

info depth 81 seldepth 68 multipv 1 score mate 34 nodes 234620749765 nps 86244184 hashfull 999 tbhits 0 time 2720424 pv h3h8 g8f7 h1f1 [SNIP]

And first mate score in less than 20 minutes:

info depth 61 seldepth 74 multipv 1 score mate 51 lowerbound nodes 93486215208 nps 82113856 hashfull 970 tbhits 0 time 1138495 pv h3h8

Regards,

Zenmastur

Exactly right, but you might as well talk to a rock. In any case, please formally release him from whatever obligation he has related to your mate challenge. It's agonizing to think about the amount of time he has wasted on it.

[mmt]

I've been doing something else in the meantime, so it will be probably another day before I have the next move.

[Uri Blass]

Of course it is not mate in 3 so 3.Rxf5 is not mate but the idea is simply to post a single line when white may have faster mate but if black does not follow the single line you then white can also mate in at most 35 moves.

But I didn't save the line, and it has been overwritten to a mate in 34 already. Some other mate in 35 line wouldn't work because Zullil wants to see the one that I found in 1.5 days, this mate in 34 was found after 17 days. And, anyway, I'm checking these lines for correctness myself, their correctness is irrelevant because the only correctness that matters is from the line I'll send for the challenge. Zullil kind of spoiled the challenge after announcing a mate in 28 (so I know what not to send), but I have not found a line to send anyway (all the lines I've seen could be improved trivially. Who knows if I'll make it to 28.)

To prove 1.Rh8+ is a mate in 28 you only need to solve 2 positions to mate in at most 26 moves

1.Rh8+ Kf7 2.Rf1+ Qf6
1.Rh8+ Kf7 2.Rf1+ Qf5

I believe that I have a solution for the first one that has to be a correct line by my definition assuming there is a solution for the second line because black cannot delay the mate to more than 28 moves by choosing the second line.

This whole thing has become rather ridiculous. Why Ovyron is even continuing is beyond me. At the pace he's going, most of us will be dead before he even reaches mate in 30. Much less mate in 28. Which, by the way, is not minimal. At least based on a private message from someone I trust ...

17 days to get to mate-in-34 is beyond terrible. This would take Stockfish maybe eight hours on a strong system, probably with no human guidance at all.

I was way off. About 45 minutes to get a mate-in-34, with no guidance and a full PV (not shown):

info depth 81 seldepth 68 multipv 1 score mate 34 nodes 234620749765 nps 86244184 hashfull 999 tbhits 0 time 2720424 pv h3h8 g8f7 h1f1 [SNIP]

And first mate score in less than 20 minutes:

info depth 61 seldepth 74 multipv 1 score mate 51 lowerbound nodes 93486215208 nps 82113856 hashfull 970 tbhits 0 time 1138495 pv h3h8

I think that the reason for human guide analysis is to find if there are bugs in stockfish(the only bug that I could find so far that stockfish fixed is related to multi-pv analysis.

Basically a claim of mate in 28 or shorter mate (and I understood 28 is not the shortest) is not something that we can trust because of hash collisions.

It may be interesting to see some tree when every mate in 9 inside the tree can be verified by a mate solver so you do not need to continue with black replies after finding a mate in 9 and every defence by black is included in the tree.

[Uri Blass]

I can add that another interesting question is if team of human+engine can beat unassisted angine analysis in finding a shorter mate.

The idea is to give many positions like the mate position and do a competition between 2 humans who use the same hardware when one of them give the computer 8 hours to search for mate and the opponent also use 8 hours but is allowed to go forward and backward.

Let say for the discussion that the engine to use is some developement version of stockfish that can forget mates if you go a lot forward and backward
and the engine does not need to show a mate score in the root at the end of the analysis but the human need to be able to claim a mate based on the analysis of the engine of different lines.

[Ovyron]

Do you think you found the optimal line of play when playing Harvey?

Optimal in the sense that none of my choices had an effect on the game's outcome, if I was going to lose anyway, delaying my loss 10 or 15 moves wouldn't have made a difference to me (the conditional I sent in the end is proof: it loses faster, I just sent it because I liked how the losing position looked, with me up on material).

Of course this hangs from 1.g4 being lost, I guess we'll see if you can save it for our game.

Some people don't have huge amounts of time to waste!

What wasted time? I use X amount of time for chess, and that remains fixed. Sometimes X is used to check correspondence games, sometimes X is used to play real time chess on lichess. Now part of X has been used for your challenge.

Since the prize is getting you to play 1.g4 then I just take the challenge as part of that game, and it's not time wasted. "You could have used this time for something else" doesn't work because I'd have wasted that time playing corr chess, playing real time chess, or zooming on my chessmap.

If X time is going to be wasted on chess then it makes no difference if I'm doing those things or working on your challenge instead. That's why better hardware doesn't help, I could have been done with your challenge long ago but then the extra time would have been used for chess anyway, so what's the difference?

A different question is if GAME RESULTS would be affected, and here I don't think Uri's test would work. I have become an expert in predicting opponent's moves, so I can predict what faster hardware would play, what unassisted engines would play and even what skilled centaurs would play. In 2012 I proved it by attaching a password protected rar file with the moves I predicted my opponent would play, and he did all but 2 of them, and this was me 7 years ago, I have only gotten better and that's why I could predict most of Harvey's moves, most of mmt's moves, and will probably predict most of Zenmastur's moves when I defeat your 1.g4.

[mmt]

1. g4 d5 2. Bg2 Bxg4 3. c4 c6 4. Qb3 e6 5. Qxb7 Nd7 6. Nc3 Ne7 7. cxd5 exd5 8. d4 Rb8 9. Qa6 Rb6 10. Qd3 Ng6 11. h3 Be6 12. Nf3 Bd6 13. h4 h5 14. b3 Nf6 15. Bg5 O-O 16. e3 Re8 17. Kf1 Bg4 18. Ne1 Bb4 19. Na4 Rb8 20. Nc2 Be7 21. f3 Be6 22. Nc5 Bc8 23. Kf2 Nd7 24. Ne6 Qa5 25. Bxe7 Rxe7 26. b4 Qb6 27. Ng5 Ba6 28. Qa3 Rbe8 29. Bf1 Bxf1 30. Raxf1 Qc7 31. Qd3 a5 32. a3

[d]4r1k1/2qnrpp1/2p3n1/p2p2Np/1P1P3P/P2QPP2/2N2K2/5R1R b - - 0 1
LC0 at -2.53. SF at -3.41.