Lasker - Lipke; Stockfish has difficulty after 39. g4

[peter]

By the way, this method of analyzing (going down a path to get a more accurate eval into the hash table) isn't always foolproof.

No single one method is, but Backward is most of the times more foolproof than standalone- computing only and sometimes even more then Forward only (of one or two or some engines) and now and then it prooves the other two methods fool, e.g. here.

[zullil]

One of the real points of this thread is that Stockfish at depth 78, and using the full 6 piece EGTB has a blunder on ply 4 in the PV. The move 41. h5? changes the position from a win to a draw, but SF sometimes thinks the variation with 41. h5 evaluates as high as +2.90.

The problem for SF is that while analyzing the position after 39. g4 it thinks this position (drawn)…

[d]8/1pp1nk2/p2p3p/6PP/4KP2/8/1P1B4/8 b - - 0 42

...is better than this one (winning):

[d]8/1pp1nk2/p2p3p/5PP1/4K2P/8/1P1B4/8 b - - 0 42

[EDIT] Now I'm a bit confused, since in your original post you gave Nd5 but here I see a knight on e7. My response below was based on your original post.

Even at the moment of truth, i.e., after 39. g4 Kf7 40. g5 Nd5, Stockfish takes a long time to switch from h5 to Ke4.

[d]8/1pp2k2/p2p3p/3n1KP1/5P1P/8/1P1B4/8 w - - 1 41

info depth 68 seldepth 98 multipv 1 score cp 132 nodes 66263308894 nps 40576212 hashfull 999 tbhits 533060909 time 1633058 pv h4h5 d5e7 f5e4 c7c6 d2c3 d6d5 e4e5 h6g5 f4g5 c6c5 c3d2 d5d4 e5e4 f7e6 d2f4 b7b5 f4c7 e7f5 h5h6 f5e7 c7d8 e7g6 d8b6 d4d3 e4d3 c5c4 d3e4 e6f7 b6d4 f7g8 e4f5 g8h7 d4c3 g6h4 f5f6 h4g6 f6f7 g6h4 c3e1 h4g6 e1d2 g6h8 f7f6 h8g6 d2c3 g6h4 c3e1 h4g6 f6f7 g6e5 f7e6 e5g6 e1a5 g6f4 e6f5 f4g6 a5b4 g6h4 f5e4 h4g6 e4d4 h7g8 b4d2 g8h7 d4d5 g6h4 d5e4 h4g6 d2b4 h7g8 e4f5 g8h7 f5g4 h7g8 b4c3 g8h7 g4f3 g6h4 f3e3 h4g6 c3e1 g6h8 e3d4 h8f7 e1d2 h7g6 d4d5 g6f5 d2c1 f5g6 c1e3 g6h7 d5e6 f7h8
info depth 69 currmove h4h5 currmovenumber 1
info depth 69 currmove g5h6 currmovenumber 2
info depth 69 currmove f5e4 currmovenumber 3
info depth 69 seldepth 107 multipv 1 score cp 142 lowerbound nodes 85733043050 nps 41146376 hashfull 999 tbhits 682169796 time 2083611 pv f5e4
info depth 68 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 151 lowerbound nodes 88887667120 nps 41223112 hashfull 999 tbhits 697282275 time 2156258 pv f5e4
info depth 67 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 166 lowerbound nodes 91791824752 nps 41346216 hashfull 999 tbhits 709899738 time 2220078 pv f5e4
info depth 66 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 186 lowerbound nodes 96401950465 nps 41490490 hashfull 999 tbhits 730964123 time 2323471 pv f5e4
info depth 65 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 213 lowerbound nodes 100566593106 nps 41612069 hashfull 999 tbhits 748175298 time 2416765 pv f5e4
info depth 64 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 250 lowerbound nodes 108507870992 nps 41783679 hashfull 999 tbhits 788078782 time 2596896 pv f5e4
info depth 63 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 298 lowerbound nodes 116571064926 nps 42008983 hashfull 999 tbhits 822905367 time 2774908 pv f5e4
info depth 62 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 361 lowerbound nodes 127379814055 nps 42247699 hashfull 999 tbhits 872514143 time 3015071 pv f5e4
info depth 61 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 441 lowerbound nodes 141629273788 nps 42528580 hashfull 999 tbhits 929968783 time 3330214 pv f5e4
info depth 60 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 543 lowerbound nodes 160220193485 nps 42934849 hashfull 999 tbhits 1008041727 time 3731705 pv f5e4
info depth 59 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 674 lowerbound nodes 182627774409 nps 43451797 hashfull 999 tbhits 1092549723 time 4202997 pv f5e4
info depth 58 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 838 lowerbound nodes 217547089714 nps 44256683 hashfull 999 tbhits 1239137122 time 4915576 pv f5e4
info depth 57 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 1047 lowerbound nodes 286285181065 nps 45646094 hashfull 999 tbhits 1588491918 time 6271844 pv f5e4
info depth 56 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 1309 lowerbound nodes 460742003116 nps 47808863 hashfull 999 tbhits 2697788307 time 9637167 pv f5e4
info depth 55 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 1639 lowerbound nodes 869920994431 nps 50325123 hashfull 999 tbhits 5314440857 time 17286018 pv f5e4
info depth 54 currmove f5e4 currmovenumber 1
info depth 69 seldepth 107 multipv 1 score cp 2054 lowerbound nodes 1718130466921 nps 51349001 hashfull 999 tbhits 12356035159 time 33459861 pv f5e4

[Vinvin]

What's the conclusion about this position ?

[Morten Lohne]

So really, the point is that sf plays the wrong move in the position after 39...Kf7 40.g5 Nd5? ( 8/1pp2k2/p2p3p/3n1KPP/5P2/8/1P1B4/8 b - - 0 41 )

Are you sure 41.h5 is drawn though? Or that 41.Ke4 is truly winning? It's a spectacularly complicated position, and +3 doesn't mean it's won.

[TommyTC]

Now I'm a bit confused, since in your original post you gave Nd5 but here I see a knight on e7. My response below was based on your original post.

Here is what I did....
I set up the position after 39. g4.
Analyzed for 12 hours to depth 78.
Looked at the PV and eval:
39...Kf7 40.g5 Nd5 41.h5 Ne7 42.Ke4
+- (2.29 --) Depth: 78/104

But the move 41. h5?? changes the position from a win to a draw. It is a bad move early in the very long PV.

On the other hand, the following variation is winning:
39...Kf7 40.g5 Nd5 41.Ke4! Ne7 42.f5!

Note that 42. h5?? can be played in this variation to arrive at the same position as in the "bad" PV variation above, but the move 42. f5! is the winning move. Note also the only difference in the 2 positions is the placement of the f and h pawns.

If you set up the position after 39...Kf7 40. g5 Nd5 41. Ke4 Ne7, then SF finds 42. f5! as the best (and winning) move rather quickly.

It is surprising to me that SF makes a huge mistake on ply 4 of a 78 ply search.

[zullil]

It is surprising to me that SF makes a huge mistake on ply 4 of a 78 ply search.

The only move in a PV that is reliable is the first one, and then only sometimes.