16.f3 - Draw? Deep Blue vs Fritz (1995)

[zullil]

todays programs instantly see c4 is a 5 point blunder. It would seem this move is the result of a bug, it was found and repaired. Programs don't blunder but 2800 GM blunder several times in every tournament

16.c4? - Mate in 34 moves is not my question.

Question is can black win after 16.f3?
After 45.Qe1 - stockfish show evaluation -0.98 from D=1 to D=67.
Neither progress, nor eval drop.

Edit: Stockfish seems to prefer Rh6 immediately after f3:

Code: Select all

info depth 34 seldepth 56 multipv 1 score cp 76 lowerbound nodes 14650370915 nps 22991324 hashfull 999 tbhits 0 time 637213 pv g6h6
info depth 34 currmove g6h6 currmovenumber 1
info depth 34 seldepth 56 multipv 1 score cp 96 lowerbound nodes 14937531585 nps 22964919 hashfull 999 tbhits 0 time 650450 pv g6h6
info depth 34 currmove g6h6 currmovenumber 1
info depth 34 seldepth 56 multipv 1 score cp 124 lowerbound nodes 16671047726 nps 23062112 hashfull 999 tbhits 0 time 722876 pv g6h6

Code: Select all

info depth 34 seldepth 63 multipv 1 score cp 129 nodes 20862098472 nps 23239654 hashfull 999 tbhits 0 time 897694 pv g6h6 d1e1 d8g5 e1f2 f8e7 d5e7 e8e7 c2c3 c8g8 f1d1 g5h5 f2g1 h6g6 d1d2 e6h3 d3f1 h5f3 a3c2 h7h5 c2e1 f3e4 g1f2 f4f3 f2f3 e4f3 e1f3 h3e6 a2a4 c6a5 d2f2 a5b3 a1a3 e5e4 f3h4 g6f6 f2f6 e7f6 a4b5 b3d2 a3a1 f6g5 b5a6 g5h4 f1b5 d2b3 a1a3
info depth 35 currmove g6h6 currmovenumber 1
info depth 35 seldepth 63 multipv 1 score cp 136 lowerbound nodes 21971675045 nps 23293882 hashfull 999 tbhits 0 time 943238 pv g6h6
info depth 35 currmove g6h6 currmovenumber 1
info depth 35 seldepth 63 multipv 1 score cp 143 lowerbound nodes 23081818258 nps 23365711 hashfull 999 tbhits 0 time 987850 pv g6h6
info depth 35 currmove g6h6 currmovenumber 1
info depth 35 seldepth 63 multipv 1 score cp 153 lowerbound nodes 26755901094 nps 23505222 hashfull 999 tbhits 0 time 1138296 pv g6h6

Code: Select all

info depth 43 seldepth 73 multipv 1 score cp 164 nodes 229713367161 nps 25123435 hashfull 999 tbhits 0 time 9143390 pv g6h6 d1e1 d8g5 e1f2 f8e7 d5e7 e8e7 c2c3 c8g8 f1d1 e6d7 a1c1 g5h5 f2g1 h6g6 d1d2 d7h3 d3f1 h5f3 a3c2 e7f8 c2e1 f3e4 g1f2 e4e3 f2e3 f4e3 d2e2 h3e6 e2e3 f8e7 c1a1 f7f5 a2a4 f5f4 e3f3 e6c4 a4b5 a6b5 h1g1 e7e6 f3h3 g8g7 a1a6 c6e7 a6a7 e5e4 h3h4 e6e5 f1c4 b5c4 g1f2 d6d5 h4h5 e7f5 a7g7

Code: Select all

info depth 45 seldepth 73 multipv 1 score cp 176 nodes 337897462690 nps 25220598 hashfull 999 tbhits 0 time 13397678 pv g6h6 f1f2 d8h4 d1g1 f8e7 c2c3 e7d8 g2g3 f4g3 f2g2 c6e7 g2g3 e7d5 e4d5 e6d5 g1g2 d8b6 a3c2 a6a5 a1f1 e8e7 a2a3 d5c6 d3f5 c8b8 f5d3 b8d8 d3f5 h4f4 f5e4 c6d7 e4d3 h6h5 f1e1 d8b8 e1f1 b6a7 f1e1 d7e6 e1f1 h5h4 f1e1 e6c4 d3f1 c4d5 f1d3 d5e6 g3g7 h4h2 g2h2

[yurikvelo]

Rh6 was pruned beacause 19. Nxe7 Kxe7 has close to 0 eval up to D=23 (D=28 from 16. .. Rh6).
It fails low at D=24 (D=29 from 16. .. Rh6) when K9.3 find 19. Nxe7 Nxe7

At D=29 and 30 (D=34 and 35 from 16. .. Rh6) 19. Nxe7 is better than 19. c3 (-0.79 vs -1.23)

[zullil]

Rh6 was pruned beacause 19. Nxe7 Kxe7 has close to 0 eval up to D=23 (D=28 from 16. .. Rh6).
It fails low at D=24 (D=29 from 16. .. Rh6) when K9.3 find 19. Nxe7 Nxe7

At D=29 and 30 (D=34 and 35 from 16. .. Rh6) 19. Nxe7 is better than 19. c3 (-0.79 vs -1.23)

Well, Stockfish hasn't (yet) seen any reason to prune it here:

Code: Select all

info depth 46 seldepth 82 multipv 1 score cp 190 nodes 514320667330 nps 25386683 hashfull 999 tbhits 0 time 20259467 pv g6h6 f1f2 d8h4 d1g1 f8e7 d5e7 c6e7 c2c3 e7g8 g2g4 f4g3 f2g2 g8f6 g2g3 e8e7 g3g2 f6h5 d3f1 h5f4 g2g3 d6d5 e4d5 e6d5 a3c2 f4h5 g3h3 h4h3 g1g5 f7f6 f1h3 d5f3 h1g1 f6g5 h3c8 h5f4 c2e3 h6f6 g1f2 f4d3 f2g1 d3b2 a1f1 e7f7 c8a6 f6a6 f1f3 f7g6 f3f2 b2d3 f2e2 a6a3 e2c2 e5e4 g1f1 h7h6 e3d5 g6f5 d5c7 a3a5 f1e2 b5b4 c3b4 d3b4 c2b2 a5a2 b2a2 b4a2 e2e3 a2b4 c7b5 b4c2 e3d2

[bob]

I think that it was not a bug and deep blue prototype did not see deep enough.


Probably it only saw
16.c4 Qh4 17.g3 Qh3 18.Rg1 Qxh2+ 19.Kxh2 Rh6+ 20.Kg2 Bh3+ 21.Kh2 Bf1+ 22.Qh5 when the fact that 22...Rxh5 is mate was too deep for it.

Here is houdini3's analysis after 16.c4 when I told it to search only for mate in 6 so 22...Rxh5 mate is too deep for it.

FEN: 2rqkb2/5p1p/p1npb1r1/1p1Np3/2P1Pp2/N2B4/PP3PPP/R2Q1R1K b - c3 0 16

Houdini_3_x64:
4/13 00:00 4k 1,495k -0.12 16. ... Nd4 17.cxb5 Qg5 18.g3
4/13 00:00 8k 917k +0.23 16. ... Qg5 17.g3 Rh6 18.cxb5 axb5 19.gxf4
5/13 00:00 10k 950k +0.23 16. ... Qg5 17.g3 Rh6 18.cxb5 axb5 19.gxf4
6/13 00:00 22k 945k +0.10 16. ... Qg5 17.g3 Rh6 18.cxb5 Nd4 19.Qa4 Bxd5 20.exd5
7/13+ 00:00 24k 978k +0.18 16. ... Qg5
7/13+ 00:00 25k 978k +0.34 16. ... Qg5
7/13 00:00 32k 999k +0.11 16. ... Qg5 17.g3 Rh6 18.cxb5 fxg3 19.fxg3 Nd4 20.Kg1 Bxd5 21.exd5 Qe3+ 22.Rf2
8/13 00:00 41k 994k +0.11 16. ... Qg5 17.g3 Rh6 18.cxb5 fxg3 19.fxg3 Nd4 20.Kg1 Bxd5 21.exd5 Qe3+ 22.Rf2
8/13+ 00:00 46k 1,051k +0.19 16. ... Qh4
8/13+ 00:00 49k 1,097k +0.35 16. ... Qh4
8/13 00:00 75k 1,370k +0.10 16. ... Qh4 17.Be2 Bxd5 18.exd5 Nd4 19.cxb5 Qg5 20.g3 fxg3 21.hxg3
8/13 00:00 75k 1,370k +0.11 16. ... Qg5 17.g3 Rh6 18.cxb5 fxg3 19.fxg3 Nd4 20.Kg1 Bxd5 21.exd5 Qe3+ 22.Rf2
9/13+ 00:00 134k 1,740k +0.19 16. ... Qh4
9/13+ 00:00 136k 1,623k +0.35 16. ... Qh4
9/13 00:00 150k 1,558k +0.48 16. ... Qh4 17.cxb5 Rh6 18.Nf6+ Kd8 19.h3 Bxh3 20.Kg1 Be6 21.f3 Qh2+ 22.Kf2
10/13+ 00:00 184k 1,602k +0.56 16. ... Qh4
10/13+ 00:00 193k 1,546k +0.74 16. ... Qh4
10/13+ 00:00 204k 1,497k +1.21 16. ... Qh4
10/13 00:00 275k 1,664k +1.22 16. ... Qh4 17.g3 Qh3 18.Rg1 Qxh2+ 19.Kxh2 Rh6+ 20.Qh5 Rxh5+ 21.Kg2 fxg3 22.Kxg3
11/13- 00:00 292k 1,688k +1.11 16. ... Qh4 17.g3
11/13- 00:00 317k 1,749k +0.90 16. ... Qh4 17.g3
11/13 00:00 341k 1,756k +0.51 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.cxb5 Bxd5 21.exd5 Nd4 22.fxg3
12/13- 00:00 388k 1,821k +0.43 16. ... Qh4 17.g3
12/13- 00:00 522k 2,014k +0.28 16. ... Qh4 17.g3
12/13 00:00 661k 2,210k +0.25 16. ... Qh4 17.g3 fxg3 18.fxg3 Rxg3 19.Qe1 Rh3 20.Qxh4 Rxh4 21.cxb5 axb5 22.Nxb5
13/13- 00:00 775k 2,361k +0.17 16. ... Qh4 17.g3
13/13- 00:00 1,091k 2,666k +0.01 16. ... Qh4 17.g3
13/13 00:00 1,488k 2,999k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
14/13 00:00 1,704k 3,109k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
15/13 00:00 2,431k 3,482k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
16/13 00:00 3,446k 3,693k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
17/13 00:01 4,868k 3,931k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
18/13 00:01 7,508k 4,339k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
19/13 00:02 10,870k 4,563k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
20/13 00:03 17,625k 4,845k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
21/13 00:05 25,800k 5,086k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
22/13 00:07 37,669k 5,280k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
23/13 00:09 52,927k 5,471k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
24/13 00:13 74,647k 5,643k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
25/13 00:17 103,336k 5,803k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
26/13 00:23 138,363k 5,875k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
27/13 00:30 185,230k 5,979k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
28/13 00:41 250,519k 6,044k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
29/13 00:54 335,822k 6,119k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
30/13 01:11 443,767k 6,199k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6
31/13 01:34 587,238k 6,237k 0.00 16. ... Qh4 17.g3 Qh3 18.Rg1 Rh6 19.Rg2 fxg3 20.fxg3 bxc4 21.Bxc4 Bg7 22.Bxa6

Here is the analysis when I allow it to search for mate in 7 so it see that 17.Rg1 is bad

FEN: 2rqkb2/5p1p/p1npb1r1/1p1Np3/2P1Pp2/N2B4/PP3PPP/R2Q1R1K b - c3 0 16

Houdini_3_x64:
4/15 00:00 5k 1,506k -0.12 16. ... Nd4 17.cxb5 Qg5 18.g3
4/15 00:00 8k 1,205k +0.23 16. ... Qg5 17.g3 Rh6 18.cxb5 axb5 19.gxf4
5/15 00:00 10k 1,212k +0.23 16. ... Qg5 17.g3 Rh6 18.cxb5 axb5 19.gxf4
6/15 00:00 17k 1,153k +0.27 16. ... Qg5 17.Rg1 Qh4 18.Rf1 Bxd5 19.exd5 Rh6
7/15+ 00:00 21k 1,223k +0.35 16. ... Qg5
7/15 00:00 36k 1,241k +0.10 16. ... Qg5 17.g3 Rh6 18.cxb5 axb5 19.Nxb5 fxg3 20.fxg3 Qxg3 21.Nbc7+ Kd7 22.Nxe6 fxe6 23.Rf7+
8/15+ 00:00 42k 1,267k +0.18 16. ... Qg5
8/15+ 00:00 43k 1,295k +0.34 16. ... Qg5
8/15 00:00 76k 1,405k +0.05 16. ... Qg5 17.g3 Ne7 18.Kg1 Bg4 19.Be2 Bh3 20.Bh5 Bxf1 21.Bxg6 hxg6 22.Kxf1 Nxd5 23.Qxd5
8/15 00:00 104k 1,605k +0.63 16. ... Qh4 17.b4 Bxd5 18.exd5 Rh6 19.h3 Nxb4 20.Rb1 Nxd3 21.Qxd3
9/15+ 00:00 118k 1,655k +0.73 16. ... Qh4
9/15+ 00:00 124k 1,677k +0.94 16. ... Qh4
9/15+ 00:00 148k 1,808k +1.43 16. ... Qh4
9/15 00:00 177k 1,903k +1.35 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Bxf1 fxg3 23.hxg3
10/15- 00:00 191k 1,928k +1.24 16. ... Qh4 17.g3
10/15 00:00 229k 1,911k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
11/15 00:00 271k 2,049k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
12/15 00:00 353k 2,246k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
13/15 00:00 555k 2,580k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
14/15 00:00 853k 3,013k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
15/15 00:00 1,356k 3,236k +1.18 16. ... Qh4 17.g3 Qh3 18.Qf3 Bg4 19.Qg2 Qh5 20.Nxb5 axb5 21.cxb5 f3 22.Qg1 Nd4 23.b6
16/15 00:00 2,032k 3,590k +1.18 16. ... Qh4 17.g3 Qh3 18.Qf3 Bg4 19.Qg2 Qh5 20.Nxb5 axb5 21.cxb5 f3 22.Qg1 Nd4 23.b6
17/15 00:00 3,241k 3,904k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
18/15 00:01 6,940k 4,364k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
19/15 00:02 10,345k 4,575k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
20/15 00:03 16,968k 4,812k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
21/15 00:05 25,152k 4,979k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
22/15 00:07 37,796k 5,130k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
23/15 00:10 55,934k 5,273k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
24/15 00:15 82,639k 5,408k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
25/15 00:22 123,809k 5,479k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
26/15 00:31 175,981k 5,536k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
27/15 00:45 255,828k 5,604k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
28/15 01:05 369,933k 5,683k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
29/15 01:29 521,238k 5,793k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
30/15 02:02 713,135k 5,822k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
31/15 02:48 988,808k 5,882k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3
32/15 03:47 1,350,410k 5,948k +1.16 16. ... Qh4 17.g3 Qh3 18.Qf3 Rh6 19.Qg2 Qxg2+ 20.Kxg2 Bh3+ 21.Kg1 Bxf1 22.Rxf1 fxg3 23.fxg3



Note that when I looked at the game I thought that maybe deep blue did not extend the line 16...Qh4 17.g3 Qh3 18.Rg1 Qxh2+ because black wasted a tempo by the moves 16...Qh4 17...Qh3 18...Qxh2+ and maybe at that time it had a rule not to extend lines when you waste a tempo.

Of course deep blue did not blunder by 18.Rg1 because at that time the line was not too deep for it.

It was all caused by a well-known problem. They had predicted the correct move a few moves back, but the network link from Hong Kong failed and they had to re-start. And they then incorrectly used too little move since they were starting from scratch.

You can find this in write-ups about the game. A little short on depth because they searched a little less time than normal, because of a network failure. Stuff happens like this all the time.