Winning after Qf6+!

[peter]

1. Qf6+ Bxf6 2. gxf6+ Kg8 3. Ng5 Nxa3 4. Rxh7 Nb5 5. h4 Nxd4

[d]1r3rk1/5p1R/1qb1pPp1/3pP1N1/p1pn3P/1pP5/1P2B3/3RB2K w - - 0 6
Stockfish-dev says:

+9.24 6. Rg7+ Kh8 7. cxd4 c3 8. Bxc3 Bb5 9. Rh7+ Kg8 10. Bg4 Qa6 11. Rg7+ Kh8 12. Rd2 a3 13. h5 a2 14. hxg6 a1=Q+ 15. Bd1 Qxd1+ 16. Rxd1 fxg6 17. Bb4 Bd3 18. Kg2 Bf1+ 19. Rxf1 Qe2+ 20. Rf2 Qh5 21. Rh7+ Qxh7 22. Nxh7 Kxh7 23. Bxf8 Rxf8 24. Kh3 Rc8 25. Rf3 Kg8 26. Kg4 Rc2 27. Rxb3 Rg2+ 28. Kf3 Rh2 29. Rb8+ Kf7 30. Rb7+ Kf8 31. Re7 Rh4 32. Ke3 g5 33. Rxe6 Rh3+ 34. Kd2 Rh4 35. Kd3 Rh3+ 36. Kc2 Re3 37. Kd2 Rf3 38. Re7 g4 39. Rg7 Ke8 40. Rxg4 Kd8 41. Ke2 Rf5 42. Rg8+ Kc7 43. Rg7+ Kc8 44. Ke3 (depth 53, 0:06:40)

Thanks, Louis, stored it too.
After another quick Backward:

1r3rk1/5p1R/1qb1pPp1/3pP1N1/p1pn3P/1pP5/1P2B3/3RB2K w - - 0 1

Analysis by asmFishWCP_2019-10-11_popcnt:

27.Tg7+ Kh8 28.cxd4 c3 29.Lxc3 Lb5 30.Lg4 Da6 31.Td2 a3 32.h5 a2 33.Th7+ Kg8 34.hxg6 a1D+ 35.Ld1 Dxd1+ 36.Txd1 fxg6 37.Tg7+ Kh8 38.Lb4 Ld3 39.Kg2 Lf1+ 40.Txf1 De2+ 41.Tf2 Dg4+ 42.Kf1 Dd1+ 43.Le1 Dd3+ 44.Kg2 De3 45.Ld2 Dxd4 46.Th7+ Kg8 47.Kf1 Dxe5 48.Tg7+ Kh8 49.Txg6 Ta8 50.f7 d4 51.Th6+ Kg7 52.Sxe6+ Dxe6 53.Txe6 Txf7 54.Tb6 Te7 55.Txb3 Ta1+ 56.Kg2 Te6 57.Tg3+ Tg6 58.Txg6+ Kxg6 59.Kg3 Ta6 60.b4 Te6 61.Tf4 d3 62.Tf2 Te4 63.Tf3 Tc4 64.Kf2 Te4 65.Txd3 Th4 66.Td8 Th3 67.Td4 Tb3 68.Ke2 Kf7 69.Td3 Tb1 70.Kf3 Kf6 71.Ke4 Ke6 72.Kd4 Kd6 73.Kc4+ Kc6 74.Td4 Kb7 75.Lc3 Tc1
+- (5.68) Tiefe: 51/98 00:02:00 3284MN

At least some more Rg7 Kh8 Rh7 Kg8 can be spared still.


So 5...Nxd4 (?) yet looses too.
Will have a closer look at 5...Nd6 even more now, if it really holds, (and I haven't found a way to avoid the fortress till now) I won't come back to this, and I just take back the ?! for it in my .pgn too.

[jp]

1r3rk1/5p1R/1qb1pPp1/3pP1N1/p1pn3P/1pP5/1P2B3/3RB2K w - - 0 1

Analysis by asmFishWCP_2019-10-11_popcnt:

27.Tg7+ Kh8 28.cxd4 c3 29.Lxc3 Lb5 30.Lg4 Da6 31.Td2 a3 32.h5 a2 33.Th7+ Kg8 34.hxg6 a1D+ 35.Ld1 Dxd1+ 36.Txd1 fxg6 37.Tg7+ Kh8 38.Lb4 Ld3 39.Kg2 Lf1+ 40.Txf1 De2+ 41.Tf2 Dg4+ 42.Kf1 Dd1+ 43.Le1 Dd3+ 44.Kg2 De3 45.Ld2 Dxd4 46.Th7+ Kg8 47.Kf1 Dxe5 48.Tg7+ Kh8 49.Txg6 Ta8 50.f7 d4 51.Th6+ Kg7 52.Sxe6+ Dxe6 53.Txe6 Txf7 54.Tb6 Te7 55.Txb3 Ta1+ 56.Kg2 Te6 57.Tg3+ Tg6 58.Txg6+ Kxg6 59.Kg3 Ta6 60.b4 Te6 61.Tf4 d3 62.Tf2 Te4 63.Tf3 Tc4 64.Kf2 Te4 65.Txd3 Th4 66.Td8 Th3 67.Td4 Tb3 68.Ke2 Kf7 69.Td3 Tb1 70.Kf3 Kf6 71.Ke4 Ke6 72.Kd4 Kd6 73.Kc4+ Kc6 74.Td4 Kb7 75.Lc3 Tc1
+- (5.68) Tiefe: 51/98 00:02:00 3284MN

So 5...Nxd4 (?) yet looses too.

Peter, is it easy to see where this line deviates from the line you previously had in this position that seemed to draw? (I don't think you posted it.)

[peter]

Peter, is it easy to see where this line deviates from the line you previously had in this position that seemed to draw? (I don't think you posted it.)

I didn't. I made the once and always main mistake to look at an output only without playing it out and having some Forward Backward with it.

The one above again is an output after backwarding the output- line of Louis' SF, so the engine had the time not only for stand alone computing (which sometimes simply has to be much longer and sometimes prunes away hot shots in big depthes or even of a few plies off only),but started pondering all over again with already full hash from the zugzwang- points along the plies later on.

In the meantime I gave the Nd6- line of Louis' SF, leading to a fortress, some Backward with Honey X6 FD, the fortress detection code from Joseph Ellis seems to work here really fine, Honey keeps the 0.00 in hash back to the root position after 26...Nd6.
Yet the output doesn't show much because of early repetition of Rg8 Kh7 Rh7 Kg8.
Yet that's the main trap for the engines here still, to take such threefolds for granted too early.
Probably that will have been the case with the first output after Nd4 (?) too.

[zullil]

In the meantime I gave the Nd6- line of Louis' SF, leading to a fortress, some Backward with Honey X6 FD, the fortress detection code from Joseph Ellis seems to work here really fine, Honey keeps the 0.00 in hash back to the root position after 26...Nd6.
Yet the output doesn't show much because of early repetition of Rg8 Kh7 Rh7 Kg8.
Yet that's the main trap for the engines here still, to take such threefolds for granted too early.
Probably that will have been the case with the first output after Nd4 (?) too.

[d]1r3rk1/5p1p/1qb1pPp1/3pP3/p1pP4/PpP2N1R/1Pn1B2P/3RB2K w - - 1 3
OK, so we have decided that Ng5 only leads to a draw, while Bd2 appears to win.

[Master Om]

Thats Why Crystal which is my fav engine now is stuck at 0.00 in Ng5 line.
[pgn][Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "?"] [Black "?"] [Result "*"] [Analysis "Crystal x64"] [AnalysisTime "270"] [FEN "1r3r2/4bpkp/1qb1p1p1/3pP1P1/p1pP1Q2/PpP2N1R/1Pn1B2P/3RB2K w - - 0 1"] [SetUp "1"] 1. Qf6+ Bxf6 2. gxf6+ Kg8 3. Ng5 h5 4. Bxh5 Rb7 5. Bf2 Nb4 6. cxb4 c3 7. Rxc3 Be8 8. Rh3 Qc6 9. Bg4 Qc2 10. Rh7 Qxb2 11. Rg7+ Kh8 12. Rh7+ {(0:04:30) 4547147kN}{[%eval 0,36,Crystal x64][%meval 270s|TB:38]}*[/pgn]

[zullil]

Thats Why Crystal which is my fav engine now is stuck at 0.00 in Ng5 line.
[pgn][Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "?"] [Black "?"] [Result "*"] [Analysis "Crystal x64"] [AnalysisTime "270"] [FEN "1r3r2/4bpkp/1qb1p1p1/3pP1P1/p1pP1Q2/PpP2N1R/1Pn1B2P/3RB2K w - - 0 1"] [SetUp "1"] 1. Qf6+ Bxf6 2. gxf6+ Kg8 3. Ng5 h5 4. Bxh5 Rb7 5. Bf2 Nb4 6. cxb4 c3 7. Rxc3 Be8 8. Rh3 Qc6 9. Bg4 Qc2 10. Rh7 Qxb2 11. Rg7+ Kh8 12. Rh7+ {(0:04:30) 4547147kN}{[%eval 0,36,Crystal x64][%meval 270s|TB:38]}*[/pgn]

An evaluation of 0.00 in the line you've posted is wrong. 3...h5 is a blunder that allows White to win. 3...Nxa3 is correct.

See http://www.talkchess.com/forum3/viewtop ... 65#p815373

[Master Om]

Thats Why Crystal which is my fav engine now is stuck at 0.00 in Ng5 line.
[pgn][Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "?"] [Black "?"] [Result "*"] [Analysis "Crystal x64"] [AnalysisTime "270"] [FEN "1r3r2/4bpkp/1qb1p1p1/3pP1P1/p1pP1Q2/PpP2N1R/1Pn1B2P/3RB2K w - - 0 1"] [SetUp "1"] 1. Qf6+ Bxf6 2. gxf6+ Kg8 3. Ng5 h5 4. Bxh5 Rb7 5. Bf2 Nb4 6. cxb4 c3 7. Rxc3 Be8 8. Rh3 Qc6 9. Bg4 Qc2 10. Rh7 Qxb2 11. Rg7+ Kh8 12. Rh7+ {(0:04:30) 4547147kN}{[%eval 0,36,Crystal x64][%meval 270s|TB:38]}*[/pgn]

An evaluation of 0.00 in the line you've posted is wrong. 3...h5 is a blunder that allows White to win. 3...Nxa3 is correct.

See http://www.talkchess.com/forum3/viewtop ... 65#p815373

I have only ran it for few minutes unaided. That's the result. Need to run more time to see if it finds. I have not assisted the program.

[Ovyron]

Please stick to working on your depth 22 opening book, so that you can win chess games with your special moves. That seems important to you.

The point is that if Qf6 is a win, or a draw, you can find the truth about its status, and the most important mainlines (so that you win/draw against anything the opponent throws at you) without ever needing to go over depth 22. And then you'll find that those high depth lines that the engine shows after several hours will be just one of those you found withing minutes, and it'll probably have a less accurate eval than what you have.

I used to analyze like peter, playing the variations to feed the hash and then backtrack so the engine shows much more accurate evals and moves up to the root, so you get much better moves to play forward. At the time engines like Rybka had a "Preserve Analysis" option for this, but Stockfish wouldn't be able to do this, as it would just forget the analysis, so I helped develop a Stockfish version that could take advantage of future positions already on the hash. A version of the code was built that wouldn't weaken the engine at games and it made it to the official branch, and since then people assisting the engine like this are getting a great edge over those that leave the engine analyzing on the same position for hours.

Later on I found out that that's not necessary, you just need to have a third party software like Chess Openings Wizard and backsolve manually scores of positions, and then you'll have accurate scores of everything you analyze, and all you need to do for positions you already analyzed is exclude the moves you already have evaluated and ask the engine what it thinks about the rest with something like my patched McBrain X with Smarter Tactical setting, and this will beat the analysis of a hash full of relevant positions, something like Persistent Hash, or a Stockfish with learning, but you have to actually put the work into it and analyze the relevant positions to evaluate them.

It's fine that you have me in ignore now, because THIS IS NOT FOR YOU, this is for the general public, a service in case they're planning to buy some CPU yet to come that will give them some unprecedented speed for their engines, so they can leave them analyzing a position overnight and get quality results when they wake up. If what peter is doing is producing much better analysis in much shorter time then they're just wasting their money and would be better doing what I call lern2analyze, which can be done in any hardware, as long as you get to depth 22 in reasonable time...

I've seen 160,000kn/s, that's 80 times faster than what I have, and I haven't been impressed. You can get much better analysis by just taking a closer look at the lines.

[jhellis3]

An evaluation of 0.00 in the line you've posted is wrong. 3...h5 is a blunder that allows White to win. 3...Nxa3 is correct.

How sure of this line are you? Not immediately clear to me why Black sacs its queen on move 28?

I did a quick multi-pv search with Crystal after shuffling the f rook instead, and it doesn't seem find anything better than 0.00 for white?

[Master Om]

An evaluation of 0.00 in the line you've posted is wrong. 3...h5 is a blunder that allows White to win. 3...Nxa3 is correct.

How sure of this line are you? Not immediately clear to me why Black sacs its queen on move 28?

I did a quick multi-pv search with Crystal after shuffling the f rook instead, and it doesn't seem find anything better than 0.00 for white?

Same here.