Tactical endgame position- some engines get it, some not.

Discussion of chess software programming and technical issues.

Moderators: Harvey Williamson, Dann Corbit, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
OliverBr
Posts: 655
Joined: Tue Dec 18, 2007 8:38 pm
Location: Munich, Germany
Full name: Dr. Oliver Brausch
Contact:

Tactical endgame position- some engines get it, some not.

Post by OliverBr » Fri Sep 18, 2020 12:04 am

Hello, after a drawn game against a very strong opponent I detected this tactical position:

It's a draw after blunder e6. The important part is: How quick does an engine see this draw?



Now it's not trivial, because OliThink sees this draw not sooner than at depth==16. But, some engines (including Fruit2.1) see this draw as depth = 6 or even sooner.
What is the tactical trick to see this coming?

Code: Select all

 1     9      0        24  g5g1 
 2     1      0       173  g5g3 b3c4 g3g1 
 3   -68      0       505  g5g1 e6e7 g1b1 b3c3 
 4  -172      0      2038  g5g1 e6e7 g1b1 b3c2 h3h2 
 5  -197      0      4486  g5g1 e6e7 g1b1 b3c2 h3h2 e2f2 f5g4 
 6  -202      0     11642  g5g1 e6e7 g1b1 b3a2 h3h2 e2f2 f5g5 f2h2 
 7  -202      0     29857  g5g1 e6e7 g1b1 b3a2 h3h2 e2f2 f5g5 f2h2 b1b8 
 8  -205      1     63138  g5g1 e6e7 h3h2 e2f2 f5g4 f2h2 g1e1 h2g2 g4f3 g2g7 
 9  -196      6    263316  g5g3 b3c2 g3g1 e6e7 h3h2 e2f2 f5g4 f2h2 g1e1 h2h7 
10  -452      9    389756  g5g3 b3c2 g3g1 e6e7 g1g8 e7e8q g8e8 e2e8 f5f4 e8f8 f4g3 f8g8 g3f3 c2c3 
11  -443     11    444166  g5g3 b3c2 g3g1 e6e7 g1g8 e7e8q g8e8 e2e8 f5f4 e8e2 f4f3 c2d3 f3g4 
12  -440     13    553053  g5g3 b3a2 g3g1 e6e7 g1g8 e7e8q g8e8 e2e8 f5f4 e8f8 f4g3 f8g8 g3f3 g8h8 f3g4 h8f8 
13  -436     20    823399  g5g3 b3b2 g3g1 e6e7 g1g8 e7e8q g8e8 e2e8 f5f4 e8e2 f4f3 e2c2 f3f4 c2f2 f4e3 f2f5 
14  -442     64   2957538  g5g3 b3c4 g3g1 e6e7 g1c1 c4d4 c1c8 e7e8q c8e8 e2e8 f5f4 e8e3 f4g4 e3e4 g4f3 e4h4 f3g3 h4e4 
15  -384     76   3539317  g5g3 b3c4 g3g1 e6e7 g1c1 c4d4 c1c8 e7e8q c8e8 e2e8 f5g4 e8h8 g4g3 h8g8 g3f3 g8f8 f3g3 d4e3 h3h2 f8g8 g3h3 
16   -24    187   9474652  g5g1 e6e7 g1g8 e7e8q g8e8 e2e8 f5g4 e8e2 g4g3 e2e3 g3g2 b3c2 h3h2 e3e2 g2g3 e2e1 g3g2 c2d3 
Chess Engine OliThink: http://brausch.org/home/chess
OliThink GitHub:https://github.com/olithink

OliverBr
Posts: 655
Joined: Tue Dec 18, 2007 8:38 pm
Location: Munich, Germany
Full name: Dr. Oliver Brausch
Contact:

Re: Tactical positions- some engines get it, some not.

Post by OliverBr » Fri Sep 18, 2020 6:04 am

Hello again,
I would like to show a "vice versa" example, a tactical blunder two very strong engines are making. It's from the a very rare win of OliThink (5.7.9, black) against Stockfish (11, white) on medium time controls:



Stockfish/Ethereal think 32. Ra6 is best move, which is actually very big blunder. (33...Rxe3 wins the match for black). They are doing this until depth 21.
Other, "weaker/stranger" engines however refute this move immediately (e.g. Fruit, OliThink or Leela):

Here the analyze log of SF11:

Code: Select all

 21	+3.51 	1.87M  	0:00.91	Qxc5 Qxc5 Bxc5 a5 Bd4 Rg8 Rc7 Rb8 Be5 Rbf8 f3 Bh4 Ra7 Bd8 Kh2 Bf6 Rxa5 Ra8 Rb5 Bxe5+ Rxe5 Rgb8 Rb5 Rxb5 axb5 
 20	+4.29 	1.02M  	0:00.54	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rc6 Rg8 Rxa7 Rb1 Rb7 Ra8 g3 g5 Be3 Bf6 a5 h5 a6 g4 a7 
 19	+3.95 	696243	0:00.38	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rc6 Rg8 Rxa7 Rb1 Rb7 Ra1 g3 Be5 Kg2 Bf6 Be3 Be5 Rd7 Ra2 Bd2 Bf6 
 18	+3.75 	610301	0:00.34	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rc6 Rce8 Bxa7 Be5+ g3 Ra1 Be3 Ra3 b4 Rxa4 b5 Kg8 b6 Ra6 Ra7 Rxa7 bxa7 
 17	+3.90 	350067	0:00.22	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rb8 b4 h5 Raxa7 Be5+ g3 h4 f4 Re2+ Kh1 Bb2 gxh4 Bf6 b5 Kh7 Rab7 
 16	+4.11 	265230	0:00.17	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Raxa7 Rb1 b4 h5 a5 Be5+ g3 h4 f4 hxg3+ Kxg3 Rb3+ Kg4 
 15	+4.02 	159797	0:00.10	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Rf4 Raxa7 Rfe4 Be3 Rb4 Rdb7 h6 Rxb4 Bxb4 Rb7 Be7 a5 Kh7 
 14	+3.72 	116612	0:00.08	Ra6 Qb4 Bxc5 Qxc4 bxc4 Rd8 Rxd8 Rxd8 Rxa7 Bc3 Bb6 Rf8 a5 Kg8 
 13	+3.61 	97979  	0:00.06	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h5 Rc7 Be5 Rcxa7 Kh7 b4 
 12	+3.44 	50457  	0:00.04	Qxc5 Qxc5 Bxc5 Rc8 Bxa7 Rc1 Rd1 Rxd1 Rxd1 Bc3 Bb6 Kg8 Kf1 Kf7 a5 Rc8 
 11	+3.44 	30452  	0:00.02	Qxc5 Qb4 Rd3 Rg8 Qxb4 Bxb4 Rxa7 Rc8 
 10	+3.08 	23090  	0:00.02	Qxc5 Qxc5 Bxc5 Re2 Rd1 Bxf2+ Kf1 Bxc5+ Kxe2 Rf2+ Ke1 
  9	+2.86 	17639  	0:00.01	Qxc5 Qxc5 Bxc5 Re2 Bxa7 Bxf2+ Bxf2 Rexf2 Rd8 Rf5 R6d7 Rf1+ Kh2 
  8	+3.14 	10973  	0:00.01	Qxc5 Qxc5 Bxc5 Bc3 Rd3 Re1+ Kh2 Be5+ g3
  7	+3.28 	3255    	0:00.00	Ra6 Rxe3 fxe3 Qb4 Qxb4 Bf2+ Kh2 cxb4 
  6	+6.38 	1349    	0:00.00	Bxc5 Re5 Bd4 g5 
  5	+6.52 	921      	0:00.00	Bxc5 Rf5 Bxa7 
  4	+7.86 	412      	0:00.00	Ra6 Rxe3 Rxa5 Bxf2+ Kh2 
  3	+8.31 	245      	0:00.00	Ra6 g5 Rxa5 Bxa5 Qxc5 
  2	+4.21 	134      	0:00.00	Rc6 g5 
  1	+3.65 	61        	0:00.00	Bxc5 
Ethereal 12.46 is having the same problems with this position as its analysis after move 32. a6 verifies a tactical blindness for Rxe3. This is a move, that many human player can see.

Code: Select all

23	+4.53!	5.67M  	0:02.53	Rxe3! 
 23	+2.99!	5.64M  	0:02.51	Rxe3! 
 22	+0.01 	5.39M  	0:02.40	Rxe3 Rxa5 Bxf2+ Kh1 Re1+ Kh2 Bg1+ Kh1 Be3+ Kh2 Bg1+ 
 21	 -4.21 	3.83M  	0:01.71	Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rdxa7 h5 Ra8 Rg8 b4 h4 g3 hxg3+ Kxg3 Rg1+ Kh2 Rc1 Be3 Rb1 b5 Rxa8 Rxa8+ Kh7 b6 
 20	 -4.21 	1.42M  	0:00.64	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 Rff8 Raxa7 Rb8 Rab7 Rfe8 Rdc7 Rxb7 Rxb7 Kg8 b4 Ra8 a5 Bxb4 Rxb4 Rxa5 Rb7 Ra2 g3 h5 Kg2 Kh7 
 19	 -4.18 	1.15M  	0:00.52	Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Raxa7 h5 b4 Kh7 a5 g5 a6 Be5+ g3 Kg6 Rab7 g4 hxg4 hxg4 Kg2 
 18	 -4.24 	914315	0:00.42	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 Rff8 Raxa7 Rb8 Rab7 Rfe8 Rxb8 Rxb8 Bd4 Bxd4 Rxd4 Ra8 g3 Rb8 b4 Kg8 
 17	 -4.26 	737845	0:00.34	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h5 Raxa7 Rb8 Rab7 Rxb7 Rxb7 Rd5 Kf1 Rd1+ Ke2 Rg1 Kd3 Be5 a5 Rxg2 
 16	 -4.12 	484827	0:00.24	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h6 Raxa7 Rb8 Rab7 Rxb7 Rxb7 Ra5 Kf1 Kg8 Rc7 Be5 Rd7 Ra6 Ke2 Rc6 
 15	 -3.65 	314947	0:00.16	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h6 Raxa7 Rb8 Rab7 Rxb7 Rxb7 Kg8 b4 Rf7 Rb8+ Rf8 Rxf8+ Kxf8 b5 Ke7 
 14	 -3.60 	197857	0:00.11	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Re6 Rc8 Bxa7 h5 Be3 Kh7 Ree7 Rh8 b4 Rb8 Rb7 Bxb4 Rxg7+ 
 13	 -3.19 	57050  	0:00.03	Qc3 Bxc5 Qe5 Rf7 Rxf7 Qxf7 Bc3 Rxa7 
 12	 -4.32 	22607  	0:00.01	Qc3 Bxc5 Qxc4 bxc4 Rf6 Rxf6 gxf6 Bd4 a6 Bxf6+ Kg8 c5 Ba5 
 11	 -4.23 	18981  	0:00.01	Qc3 Bxc5 Qxc4 bxc4 Rf6 Raxa7 Rc6 Bd4 Rxc4 Bxg7+ 
 10	 -4.61 	15997  	0:00.01	Qc3 Bxc5 Qxc4 bxc4 Rg8 Raxa7 Bc3 Be7 Bf6 
  9	 -4.07 	2634    	0:00.00	Qc3 Raxa7 
  8	 -8.74 	1355    	0:00.00	Rd8 Rxa5 Rxd7 Rxc5 
  7	 -9.61 	939      	0:00.00	Rd8 Rxa5 Rxd7 Qxc5 Rb8 Rxa7 
  6	 -9.61 	789      	0:00.00	Rd8 Rxa5 Rxd7 Qxc5 Rb8 Rxa7 
  5	 -8.95 	451      	0:00.00	Rd8 Rxa5 Rxd7 Rxc5 Rfd8 
  4	 -8.74 	265      	0:00.00	Rd8 Rxa5 Rxd7 Rxc5 
  3	 -9.62 	121      	0:00.00	Rf5 Rxa5 Bxa5 Rxa7 
  2	 -10.00 	45        	0:00.00	Rg8 Rxa5 Bxa5 Qxc5 
  1	 -10.00 	29        	0:00.00	Rg8 Rxa5 Bxa5 Qxc5 
PS: There may be some kind of explanation: After 33...Rx3 34. Rxa5 white looks leading for an eternity, because black is check-chasing the king around the board without gaining material for a long time:

Code: Select all

39	  #29 	16.3M  	0:05.00	Bxf2+ Kh1 Re1+ Kh2 Bg1+ Kg3 Re3+ Kg4 h5+ Kg5 Rg3+ Qg4 hxg4 Raxa7 gxh3+ Kh4 h2 Kxg3 h1=Q Rf7 Bh2+ Kf2 Qg1+ Ke2 Qxg2+ Ke3 Bg1+ Kd3 Rxf7 Rxf7 Qd5+ Kc3 Bd4+ Kc2 Qxf7 a5 Qe6 b4 Qc4+ Kd2 Qxb4+ Kd3 Qxa5 Kc4 Qa4+ Kd5 Qa6 Ke4 Qc4 Kf3 Q
It seems that SF/Ethereal do not fellow check sequences too deep. This may lead to a tactical blindness, but of course gains in overall strength as such positions are rare.
Chess Engine OliThink: http://brausch.org/home/chess
OliThink GitHub:https://github.com/olithink

OliverBr
Posts: 655
Joined: Tue Dec 18, 2007 8:38 pm
Location: Munich, Germany
Full name: Dr. Oliver Brausch
Contact:

Re: Tactical endgame position- some engines get it, some not.

Post by OliverBr » Tue Sep 22, 2020 10:26 pm

I do have another fine position to discuss: Easy Mate #6, so it looks like, but both, Stockfish and Ethereal are struggling:



Mate in 6, one would expect that any engine finds it in 13 plies.
No, this is not true!

Stockfish and Ethereal need 28 plies and a lot of time for an very easy mate line, easy for human players.

OliThink 5.3.3 finds it in 13 plies. Then I started pruning and 5.8.0 needs 22 plies, though playing about 300 ELOs stronger than 5.3.3.
How can this be explained?

PS: Here Ethereals Analyzis Log, Stockfish's is not that different. Leela doesn't find it at all. Fruit, or other "weaker" engines. however find it immediately.

Code: Select all

28	    #6 	132.5M	1:06.42	Rxh7+ Kxh7 Qd3+ Kh8 Ng5 Nf6 exf6 Bf5 Qxf5 Re8 Qh7# 
 28	+304.45!	132.5M	1:06.42	Rxh7+! 
 28	+215.88!	132.5M	1:06.42	Rxh7+! 
 28	+156.83!	132.5M	1:06.42	Rxh7+! 
 28	+117.46!	132.5M	1:06.42	Rxh7+! 
 28	+91.21!	132.5M	1:06.42	Rxh7+! 
 28	+73.71!	132.5M	1:06.41	Rxh7+! 
 28	+62.04!	132.4M	1:06.41	Rxh7+! 
 28	+54.26!	132.4M	1:06.39	Rxh7+! 
 28	+49.07!	132.3M	1:06.35	Rxh7+! 
 28	+45.61!	132.2M	1:06.26	Rxh7+! 
 28	+43.30!	132.0M	1:06.17	Rxh7+! 
 28	+41.76!	131.9M	1:06.10	Rxh7+! 
 28	+40.73!	115.1M	0:57.18	Qd3! 
 28	+40.03?	113.8M	0:56.53	? 
 28	+40.97!	112.0M	0:55.61	Qd3! 
 28	+40.66!	103.9M	0:51.45	Qd3! 
 28	+40.45!	97.1M  	0:47.86	Qd3! 
 28	+40.31!	86.8M  	0:42.41	Qd3! 
 27	+40.17 	75.3M  	0:36.38	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Kf8 Rxh6 Ke8 exd6 exd6 Rxh7 Nf8 Qe4+ Be6 Nxe6 Nxh7 Qxh7 Nc6 Qxb7 Rb8 Qxc6+ Ke7 Ng5 Rd8 bxc3 a6 Bxa6 Rd7 Kf2 Ra7 Bb5 Kf8 Qxd6+ 
 27	+38.52!	62.9M  	0:30.15	Qd3! 
 27	+36.96?	62.6M  	0:30.01	? 
 27	+39.05!	62.0M  	0:29.72	Qd3! 
 27	+38.36!	61.3M  	0:29.42	Qd3! 
 27	+37.90!	60.6M  	0:29.09	Qd3! 
 27	+37.59!	59.6M  	0:28.68	Qd3! 
 27	+37.38!	58.2M  	0:28.01	Qd3! 
 27	+37.24!	56.2M  	0:27.01	Qd3! 
 26	+37.10 	55.0M  	0:26.43	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 Nc6 Qxc3+ Kg8 Rxh6 Be6 Nxe6 Nxe6 Rxe6 Kf7 Rh6 Rd8 Bc4+ Ke8 dxe7 Rd4 Rxc6 Re4+ Be2 bxc6 Qxc6+ Kxe7 Qxe4+ Kd8 Qxh7 
 26	+39.21!	52.6M  	0:25.23	Qd3! 
 26	+36.90!	43.0M  	0:20.34	Qd3! 
 26	+35.48?	42.5M  	0:20.12	? 
 26	+37.29!	40.9M  	0:19.35	Qd3! 
 26	+36.60!	37.8M  	0:17.87	Qd3! 
 26	+36.14!	34.2M  	0:16.21	Qd3! 
 26	+35.83!	29.3M  	0:13.87	Qd3! 
 26	+35.69?	27.5M  	0:12.96	? 
 26	+35.83?	26.2M  	0:12.29	? 
 25	+35.97 	24.5M  	0:11.47	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 cxb2 Bxb2 Nc6 Ne6+ Kf7 Rxh6 Ndxe5 Qxh7+ Ke8 fxe5 Bxe6 Qh8+ Kd7 Qxa8 Kc7 exd6+ exd6 Rxe6 Kb6 Rxd6 a5 Ba3 Kc7 
 25	+35.20!	20.4M  	0:09.44	Qd3! 
 25	+33.66!	16.8M  	0:07.69	Qd3! 
 25	+32.62?	16.5M  	0:07.58	? 
 25	+34.02!	16.1M  	0:07.39	Qd3! 
 25	+33.56!	15.1M  	0:06.92	Qd3! 
 25	+33.25!	13.9M  	0:06.39	Qd3! 
 25	+33.04!	12.2M  	0:05.62	Qd3! 
 25	+32.90!	10.2M  	0:04.67	Qd3! 
 24	+32.76 	9.59M  	0:04.37	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 cxb2 Bxb2 dxe5 Ne6+ Kf7 Rxh6 Nf6 Ng5+ Ke8 fxe5 Nfd7 Qxh7 Nf8 Qf7+ Kd7 Rd1+ Kc7 Qxe7+ Nbd7 Qd6+ Kd8 
 24	+32.95!	9.33M  	0:04.24	Qd3! 
 24	+32.26!	8.78M  	0:04.00	Qd3! 
 24	+31.80!	8.43M  	0:03.85	Qd3! 
 24	+31.49!	8.06M  	0:03.70	Qd3! 
 24	+31.28!	7.55M  	0:03.48	Qd3! 
 24	+31.14!	6.86M  	0:03.15	Qd3! 
 23	+31.00 	6.68M  	0:03.07	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nc5 Qxc3 Kg8 Rxh6 Nc6 exd6 Nxd8 dxe7 Bd7 exd8=Q+ Rxd8 Bc4+ Be6 Rxe6 Nxe6 Bxe6+ Kf8 f5 
 23	+32.26!	6.42M  	0:02.95	Qd3! 
 23	+31.80!	5.99M  	0:02.75	Qd3! 
 23	+31.49!	5.63M  	0:02.59	Qd3! 
 22	+31.00 	4.64M  	0:02.14	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 cxb2 Bxb2 Kf8 Rxh6 Ke8 exd6 Nc6 Ne6 Nf6 Nc7+ Kd8 Bxf6 exf6 Qxh7 Rb8 Qf7 
 21	+30.51 	2.45M  	0:01.14	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Bh3 Rxh3 Nbd7 Ne6+ Kh8 Rxh6 Rb8 bxc3 exf4 Bxf4 Nxe6 Qxe6 
 20	+29.77 	1.37M  	0:00.65	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nb6 exd6 Nc6 Nxc6 bxc6 Qxc3+ Kg8 Rxh6 Bf5 dxe7 Bg6 f5 Bxf5 Rxc6 
 19	+29.06 	834525	0:00.41	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Kh8 Rxh6 Bf5 fxe5 Nc6 Nxc6 cxb2 Bxb2 bxc6 e6+ Kg8 
 18	+28.28 	609185	0:00.31	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Bh3 Rxh3 Nbd7 Ne6+ Kh8 Rxh6 b5 Qxc3 Re8 
 17	+27.87 	490397	0:00.25	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Kf8 Rxh6 Ke8 exd6 Kxd8 dxe7+ Kxe7 Qd6+ Ke8 Rxh7 Nc6 Rh8+ Kf7 Bd3 Nf6 bxc3 
 16	+26.07 	300064	0:00.16	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Bxf4 Bxf4 Nbd7 bxc3 exf4 Qf7+ Kh8 Qxf4 Kg8 
 15	+24.97 	160288	0:00.10	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 dxe5 Ne6+ Kh8 Rxh6 Nf6 fxe5 cxb2 exf6 bxc1=Q+ Rxc1 Bxe6 fxe7 
 14	+16.65 	88400  	0:00.06	Qd3 Nf6 exf6 Bh3 fxg7+ Kxg7 Rxh3 Rxf7 Ng5 Rf6 Nxh7 cxb2 Bxb2 
 13	+13.89 	46733  	0:00.03	Ng5 Nf6 exf6 Bf5 fxg7+ Kxg7 Qh5 Nd7 Nxh7 Bxc2 Be2 Bxh7 bxc3 Rxf7 
 12	+13.49 	33361  	0:00.02	Ng5 Nf6 exf6 Bf5 fxg7+ Kxg7 Qh5 Qc8 g4 Bxc2 Nxh7 Rxf7 
 11	+12.40 	21707  	0:00.01	Ng5 Nf6 exf6 Bf5 fxg7+ Kxg7 Qd5 Qc8 Bd3 Bxd3 Qxd3 h6 
 10	+3.98 	13927  	0:00.01	Ng5 h6 Qd3 Nf6 exf6 exf6 
  9	+1.99 	11640  	0:00.01	Ng5 h6 Qd3 
  8	 -0.53 	11244  	0:00.01	Ng5 h6 
  7	+1.33 	2375    	0:00.00	Ng5 cxb2 Bxb2 dxe5 Rxh7# 
  6	 -0.59 	1169    	0:00.00	Qd5 
  5	 -1.67 	475      	0:00.00	bxc3 Rxf7 
  4	 -2.61 	257      	0:00.00	Rh5 dxe5 bxc3 e4 
  3	 -1.89 	112      	0:00.00	Rh5 Rxf7 bxc3 
  2	 -1.83 	32        	0:00.00	bxc3 dxe5 
  1	 -1.83 	16        	0:00.00	bxc3 dxe5 
  0	# 
Chess Engine OliThink: http://brausch.org/home/chess
OliThink GitHub:https://github.com/olithink

Terje
Posts: 287
Joined: Tue Nov 19, 2019 3:34 am
Location: https://github.com/TerjeKir/weiss
Full name: Terje Kirstihagen

Re: Tactical endgame position- some engines get it, some not.

Post by Terje » Tue Sep 22, 2020 10:56 pm

OliverBr wrote:
Tue Sep 22, 2020 10:26 pm
OliThink 5.3.3 finds it in 13 plies. Then I started pruning and 5.8.0 needs 22 plies, though playing about 300 ELOs stronger than 5.3.3.
How can this be explained?
You don't need to find the best move to win, you just need to find a move that is good enough. Looking for the best move will let you find this mate instantly, but will make you unable to find good enough moves in countless other positions that require you to look deeper :)

Weiss needs depth 27, with a seldepth of 40 before it finds it.

ps there's one - too many in the FEN :)

The previous puzzle was also cool, Weiss takes a while to realize Rxe3 refutes Ra6, but then immediately gives Rxg7 with a much higher score than the original line had :D

Uri Blass
Posts: 8828
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Tactical positions- some engines get it, some not.

Post by Uri Blass » Wed Sep 23, 2020 1:08 am

OliverBr wrote:
Fri Sep 18, 2020 6:04 am
Hello again,
I would like to show a "vice versa" example, a tactical blunder two very strong engines are making. It's from the a very rare win of OliThink (5.7.9, black) against Stockfish (11, white) on medium time controls:



Stockfish/Ethereal think 32. Ra6 is best move, which is actually very big blunder. (33...Rxe3 wins the match for black). They are doing this until depth 21.
Other, "weaker/stranger" engines however refute this move immediately (e.g. Fruit, OliThink or Leela):

Here the analyze log of SF11:

Code: Select all

 21	+3.51 	1.87M  	0:00.91	Qxc5 Qxc5 Bxc5 a5 Bd4 Rg8 Rc7 Rb8 Be5 Rbf8 f3 Bh4 Ra7 Bd8 Kh2 Bf6 Rxa5 Ra8 Rb5 Bxe5+ Rxe5 Rgb8 Rb5 Rxb5 axb5 
 20	+4.29 	1.02M  	0:00.54	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rc6 Rg8 Rxa7 Rb1 Rb7 Ra8 g3 g5 Be3 Bf6 a5 h5 a6 g4 a7 
 19	+3.95 	696243	0:00.38	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rc6 Rg8 Rxa7 Rb1 Rb7 Ra1 g3 Be5 Kg2 Bf6 Be3 Be5 Rd7 Ra2 Bd2 Bf6 
 18	+3.75 	610301	0:00.34	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rc6 Rce8 Bxa7 Be5+ g3 Ra1 Be3 Ra3 b4 Rxa4 b5 Kg8 b6 Ra6 Ra7 Rxa7 bxa7 
 17	+3.90 	350067	0:00.22	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rb8 b4 h5 Raxa7 Be5+ g3 h4 f4 Re2+ Kh1 Bb2 gxh4 Bf6 b5 Kh7 Rab7 
 16	+4.11 	265230	0:00.17	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Raxa7 Rb1 b4 h5 a5 Be5+ g3 h4 f4 hxg3+ Kxg3 Rb3+ Kg4 
 15	+4.02 	159797	0:00.10	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Rf4 Raxa7 Rfe4 Be3 Rb4 Rdb7 h6 Rxb4 Bxb4 Rb7 Be7 a5 Kh7 
 14	+3.72 	116612	0:00.08	Ra6 Qb4 Bxc5 Qxc4 bxc4 Rd8 Rxd8 Rxd8 Rxa7 Bc3 Bb6 Rf8 a5 Kg8 
 13	+3.61 	97979  	0:00.06	Ra6 Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h5 Rc7 Be5 Rcxa7 Kh7 b4 
 12	+3.44 	50457  	0:00.04	Qxc5 Qxc5 Bxc5 Rc8 Bxa7 Rc1 Rd1 Rxd1 Rxd1 Bc3 Bb6 Kg8 Kf1 Kf7 a5 Rc8 
 11	+3.44 	30452  	0:00.02	Qxc5 Qb4 Rd3 Rg8 Qxb4 Bxb4 Rxa7 Rc8 
 10	+3.08 	23090  	0:00.02	Qxc5 Qxc5 Bxc5 Re2 Rd1 Bxf2+ Kf1 Bxc5+ Kxe2 Rf2+ Ke1 
  9	+2.86 	17639  	0:00.01	Qxc5 Qxc5 Bxc5 Re2 Bxa7 Bxf2+ Bxf2 Rexf2 Rd8 Rf5 R6d7 Rf1+ Kh2 
  8	+3.14 	10973  	0:00.01	Qxc5 Qxc5 Bxc5 Bc3 Rd3 Re1+ Kh2 Be5+ g3
  7	+3.28 	3255    	0:00.00	Ra6 Rxe3 fxe3 Qb4 Qxb4 Bf2+ Kh2 cxb4 
  6	+6.38 	1349    	0:00.00	Bxc5 Re5 Bd4 g5 
  5	+6.52 	921      	0:00.00	Bxc5 Rf5 Bxa7 
  4	+7.86 	412      	0:00.00	Ra6 Rxe3 Rxa5 Bxf2+ Kh2 
  3	+8.31 	245      	0:00.00	Ra6 g5 Rxa5 Bxa5 Qxc5 
  2	+4.21 	134      	0:00.00	Rc6 g5 
  1	+3.65 	61        	0:00.00	Bxc5 
Ethereal 12.46 is having the same problems with this position as its analysis after move 32. a6 verifies a tactical blindness for Rxe3. This is a move, that many human player can see.

Code: Select all

23	+4.53!	5.67M  	0:02.53	Rxe3! 
 23	+2.99!	5.64M  	0:02.51	Rxe3! 
 22	+0.01 	5.39M  	0:02.40	Rxe3 Rxa5 Bxf2+ Kh1 Re1+ Kh2 Bg1+ Kh1 Be3+ Kh2 Bg1+ 
 21	 -4.21 	3.83M  	0:01.71	Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Rdxa7 h5 Ra8 Rg8 b4 h4 g3 hxg3+ Kxg3 Rg1+ Kh2 Rc1 Be3 Rb1 b5 Rxa8 Rxa8+ Kh7 b6 
 20	 -4.21 	1.42M  	0:00.64	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 Rff8 Raxa7 Rb8 Rab7 Rfe8 Rdc7 Rxb7 Rxb7 Kg8 b4 Ra8 a5 Bxb4 Rxb4 Rxa5 Rb7 Ra2 g3 h5 Kg2 Kh7 
 19	 -4.18 	1.15M  	0:00.52	Qc3 Qxc3 Bxc3 Bxc5 Re1+ Kh2 Rc8 Raxa7 h5 b4 Kh7 a5 g5 a6 Be5+ g3 Kg6 Rab7 g4 hxg4 hxg4 Kg2 
 18	 -4.24 	914315	0:00.42	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 Rff8 Raxa7 Rb8 Rab7 Rfe8 Rxb8 Rxb8 Bd4 Bxd4 Rxd4 Ra8 g3 Rb8 b4 Kg8 
 17	 -4.26 	737845	0:00.34	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h5 Raxa7 Rb8 Rab7 Rxb7 Rxb7 Rd5 Kf1 Rd1+ Ke2 Rg1 Kd3 Be5 a5 Rxg2 
 16	 -4.12 	484827	0:00.24	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h6 Raxa7 Rb8 Rab7 Rxb7 Rxb7 Ra5 Kf1 Kg8 Rc7 Be5 Rd7 Ra6 Ke2 Rc6 
 15	 -3.65 	314947	0:00.16	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Be3 h6 Raxa7 Rb8 Rab7 Rxb7 Rxb7 Kg8 b4 Rf7 Rb8+ Rf8 Rxf8+ Kxf8 b5 Ke7 
 14	 -3.60 	197857	0:00.11	Qc3 Qxc3 Bxc3 Bxc5 Rf5 Re6 Rc8 Bxa7 h5 Be3 Kh7 Ree7 Rh8 b4 Rb8 Rb7 Bxb4 Rxg7+ 
 13	 -3.19 	57050  	0:00.03	Qc3 Bxc5 Qe5 Rf7 Rxf7 Qxf7 Bc3 Rxa7 
 12	 -4.32 	22607  	0:00.01	Qc3 Bxc5 Qxc4 bxc4 Rf6 Rxf6 gxf6 Bd4 a6 Bxf6+ Kg8 c5 Ba5 
 11	 -4.23 	18981  	0:00.01	Qc3 Bxc5 Qxc4 bxc4 Rf6 Raxa7 Rc6 Bd4 Rxc4 Bxg7+ 
 10	 -4.61 	15997  	0:00.01	Qc3 Bxc5 Qxc4 bxc4 Rg8 Raxa7 Bc3 Be7 Bf6 
  9	 -4.07 	2634    	0:00.00	Qc3 Raxa7 
  8	 -8.74 	1355    	0:00.00	Rd8 Rxa5 Rxd7 Rxc5 
  7	 -9.61 	939      	0:00.00	Rd8 Rxa5 Rxd7 Qxc5 Rb8 Rxa7 
  6	 -9.61 	789      	0:00.00	Rd8 Rxa5 Rxd7 Qxc5 Rb8 Rxa7 
  5	 -8.95 	451      	0:00.00	Rd8 Rxa5 Rxd7 Rxc5 Rfd8 
  4	 -8.74 	265      	0:00.00	Rd8 Rxa5 Rxd7 Rxc5 
  3	 -9.62 	121      	0:00.00	Rf5 Rxa5 Bxa5 Rxa7 
  2	 -10.00 	45        	0:00.00	Rg8 Rxa5 Bxa5 Qxc5 
  1	 -10.00 	29        	0:00.00	Rg8 Rxa5 Bxa5 Qxc5 
PS: There may be some kind of explanation: After 33...Rx3 34. Rxa5 white looks leading for an eternity, because black is check-chasing the king around the board without gaining material for a long time:

Code: Select all

39	  #29 	16.3M  	0:05.00	Bxf2+ Kh1 Re1+ Kh2 Bg1+ Kg3 Re3+ Kg4 h5+ Kg5 Rg3+ Qg4 hxg4 Raxa7 gxh3+ Kh4 h2 Kxg3 h1=Q Rf7 Bh2+ Kf2 Qg1+ Ke2 Qxg2+ Ke3 Bg1+ Kd3 Rxf7 Rxf7 Qd5+ Kc3 Bd4+ Kc2 Qxf7 a5 Qe6 b4 Qc4+ Kd2 Qxb4+ Kd3 Qxa5 Kc4 Qa4+ Kd5 Qa6 Ke4 Qc4 Kf3 Q
It seems that SF/Ethereal do not fellow check sequences too deep. This may lead to a tactical blindness, but of course gains in overall strength as such positions are rare.
SF11 is an old version and based on your analysis it reject Ra6 in less than a second at depth 21
21 +3.51 1.87M 0:00.91 Qxc5 Qxc5 Bxc5 a5 Bd4 Rg8 Rc7 Rb8 Be5 Rbf8 f3 Bh4 Ra7 Bd8 Kh2 Bf6 Rxa5 Ra8 Rb5 Bxe5+ Rxe5 Rgb8 Rb5 Rxb5 axb5

Uri Blass
Posts: 8828
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Tactical endgame position- some engines get it, some not.

Post by Uri Blass » Wed Sep 23, 2020 1:14 am

OliverBr wrote:
Tue Sep 22, 2020 10:26 pm
I do have another fine position to discuss: Easy Mate #6, so it looks like, but both, Stockfish and Ethereal are struggling:



Mate in 6, one would expect that any engine finds it in 13 plies.
No, this is not true!

Stockfish and Ethereal need 28 plies and a lot of time for an very easy mate line, easy for human players.

OliThink 5.3.3 finds it in 13 plies. Then I started pruning and 5.8.0 needs 22 plies, though playing about 300 ELOs stronger than 5.3.3.
How can this be explained?

PS: Here Ethereals Analyzis Log, Stockfish's is not that different. Leela doesn't find it at all. Fruit, or other "weaker" engines. however find it immediately.

Code: Select all

28	    #6 	132.5M	1:06.42	Rxh7+ Kxh7 Qd3+ Kh8 Ng5 Nf6 exf6 Bf5 Qxf5 Re8 Qh7# 
 28	+304.45!	132.5M	1:06.42	Rxh7+! 
 28	+215.88!	132.5M	1:06.42	Rxh7+! 
 28	+156.83!	132.5M	1:06.42	Rxh7+! 
 28	+117.46!	132.5M	1:06.42	Rxh7+! 
 28	+91.21!	132.5M	1:06.42	Rxh7+! 
 28	+73.71!	132.5M	1:06.41	Rxh7+! 
 28	+62.04!	132.4M	1:06.41	Rxh7+! 
 28	+54.26!	132.4M	1:06.39	Rxh7+! 
 28	+49.07!	132.3M	1:06.35	Rxh7+! 
 28	+45.61!	132.2M	1:06.26	Rxh7+! 
 28	+43.30!	132.0M	1:06.17	Rxh7+! 
 28	+41.76!	131.9M	1:06.10	Rxh7+! 
 28	+40.73!	115.1M	0:57.18	Qd3! 
 28	+40.03?	113.8M	0:56.53	? 
 28	+40.97!	112.0M	0:55.61	Qd3! 
 28	+40.66!	103.9M	0:51.45	Qd3! 
 28	+40.45!	97.1M  	0:47.86	Qd3! 
 28	+40.31!	86.8M  	0:42.41	Qd3! 
 27	+40.17 	75.3M  	0:36.38	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Kf8 Rxh6 Ke8 exd6 exd6 Rxh7 Nf8 Qe4+ Be6 Nxe6 Nxh7 Qxh7 Nc6 Qxb7 Rb8 Qxc6+ Ke7 Ng5 Rd8 bxc3 a6 Bxa6 Rd7 Kf2 Ra7 Bb5 Kf8 Qxd6+ 
 27	+38.52!	62.9M  	0:30.15	Qd3! 
 27	+36.96?	62.6M  	0:30.01	? 
 27	+39.05!	62.0M  	0:29.72	Qd3! 
 27	+38.36!	61.3M  	0:29.42	Qd3! 
 27	+37.90!	60.6M  	0:29.09	Qd3! 
 27	+37.59!	59.6M  	0:28.68	Qd3! 
 27	+37.38!	58.2M  	0:28.01	Qd3! 
 27	+37.24!	56.2M  	0:27.01	Qd3! 
 26	+37.10 	55.0M  	0:26.43	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 Nc6 Qxc3+ Kg8 Rxh6 Be6 Nxe6 Nxe6 Rxe6 Kf7 Rh6 Rd8 Bc4+ Ke8 dxe7 Rd4 Rxc6 Re4+ Be2 bxc6 Qxc6+ Kxe7 Qxe4+ Kd8 Qxh7 
 26	+39.21!	52.6M  	0:25.23	Qd3! 
 26	+36.90!	43.0M  	0:20.34	Qd3! 
 26	+35.48?	42.5M  	0:20.12	? 
 26	+37.29!	40.9M  	0:19.35	Qd3! 
 26	+36.60!	37.8M  	0:17.87	Qd3! 
 26	+36.14!	34.2M  	0:16.21	Qd3! 
 26	+35.83!	29.3M  	0:13.87	Qd3! 
 26	+35.69?	27.5M  	0:12.96	? 
 26	+35.83?	26.2M  	0:12.29	? 
 25	+35.97 	24.5M  	0:11.47	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 cxb2 Bxb2 Nc6 Ne6+ Kf7 Rxh6 Ndxe5 Qxh7+ Ke8 fxe5 Bxe6 Qh8+ Kd7 Qxa8 Kc7 exd6+ exd6 Rxe6 Kb6 Rxd6 a5 Ba3 Kc7 
 25	+35.20!	20.4M  	0:09.44	Qd3! 
 25	+33.66!	16.8M  	0:07.69	Qd3! 
 25	+32.62?	16.5M  	0:07.58	? 
 25	+34.02!	16.1M  	0:07.39	Qd3! 
 25	+33.56!	15.1M  	0:06.92	Qd3! 
 25	+33.25!	13.9M  	0:06.39	Qd3! 
 25	+33.04!	12.2M  	0:05.62	Qd3! 
 25	+32.90!	10.2M  	0:04.67	Qd3! 
 24	+32.76 	9.59M  	0:04.37	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 cxb2 Bxb2 dxe5 Ne6+ Kf7 Rxh6 Nf6 Ng5+ Ke8 fxe5 Nfd7 Qxh7 Nf8 Qf7+ Kd7 Rd1+ Kc7 Qxe7+ Nbd7 Qd6+ Kd8 
 24	+32.95!	9.33M  	0:04.24	Qd3! 
 24	+32.26!	8.78M  	0:04.00	Qd3! 
 24	+31.80!	8.43M  	0:03.85	Qd3! 
 24	+31.49!	8.06M  	0:03.70	Qd3! 
 24	+31.28!	7.55M  	0:03.48	Qd3! 
 24	+31.14!	6.86M  	0:03.15	Qd3! 
 23	+31.00 	6.68M  	0:03.07	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nc5 Qxc3 Kg8 Rxh6 Nc6 exd6 Nxd8 dxe7 Bd7 exd8=Q+ Rxd8 Bc4+ Be6 Rxe6 Nxe6 Bxe6+ Kf8 f5 
 23	+32.26!	6.42M  	0:02.95	Qd3! 
 23	+31.80!	5.99M  	0:02.75	Qd3! 
 23	+31.49!	5.63M  	0:02.59	Qd3! 
 22	+31.00 	4.64M  	0:02.14	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 cxb2 Bxb2 Kf8 Rxh6 Ke8 exd6 Nc6 Ne6 Nf6 Nc7+ Kd8 Bxf6 exf6 Qxh7 Rb8 Qf7 
 21	+30.51 	2.45M  	0:01.14	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Bh3 Rxh3 Nbd7 Ne6+ Kh8 Rxh6 Rb8 bxc3 exf4 Bxf4 Nxe6 Qxe6 
 20	+29.77 	1.37M  	0:00.65	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nb6 exd6 Nc6 Nxc6 bxc6 Qxc3+ Kg8 Rxh6 Bf5 dxe7 Bg6 f5 Bxf5 Rxc6 
 19	+29.06 	834525	0:00.41	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Kh8 Rxh6 Bf5 fxe5 Nc6 Nxc6 cxb2 Bxb2 bxc6 e6+ Kg8 
 18	+28.28 	609185	0:00.31	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Bh3 Rxh3 Nbd7 Ne6+ Kh8 Rxh6 b5 Qxc3 Re8 
 17	+27.87 	490397	0:00.25	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Kf8 Rxh6 Ke8 exd6 Kxd8 dxe7+ Kxe7 Qd6+ Ke8 Rxh7 Nc6 Rh8+ Kf7 Bd3 Nf6 bxc3 
 16	+26.07 	300064	0:00.16	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 Nf8 exd6 e5 Qc4 Bxf4 Bxf4 Nbd7 bxc3 exf4 Qf7+ Kh8 Qxf4 Kg8 
 15	+24.97 	160288	0:00.10	Qd3 Bh6 Ng5 Rxf7 Nxf7+ Kg7 Nxd8 dxe5 Ne6+ Kh8 Rxh6 Nf6 fxe5 cxb2 exf6 bxc1=Q+ Rxc1 Bxe6 fxe7 
 14	+16.65 	88400  	0:00.06	Qd3 Nf6 exf6 Bh3 fxg7+ Kxg7 Rxh3 Rxf7 Ng5 Rf6 Nxh7 cxb2 Bxb2 
 13	+13.89 	46733  	0:00.03	Ng5 Nf6 exf6 Bf5 fxg7+ Kxg7 Qh5 Nd7 Nxh7 Bxc2 Be2 Bxh7 bxc3 Rxf7 
 12	+13.49 	33361  	0:00.02	Ng5 Nf6 exf6 Bf5 fxg7+ Kxg7 Qh5 Qc8 g4 Bxc2 Nxh7 Rxf7 
 11	+12.40 	21707  	0:00.01	Ng5 Nf6 exf6 Bf5 fxg7+ Kxg7 Qd5 Qc8 Bd3 Bxd3 Qxd3 h6 
 10	+3.98 	13927  	0:00.01	Ng5 h6 Qd3 Nf6 exf6 exf6 
  9	+1.99 	11640  	0:00.01	Ng5 h6 Qd3 
  8	 -0.53 	11244  	0:00.01	Ng5 h6 
  7	+1.33 	2375    	0:00.00	Ng5 cxb2 Bxb2 dxe5 Rxh7# 
  6	 -0.59 	1169    	0:00.00	Qd5 
  5	 -1.67 	475      	0:00.00	bxc3 Rxf7 
  4	 -2.61 	257      	0:00.00	Rh5 dxe5 bxc3 e4 
  3	 -1.89 	112      	0:00.00	Rh5 Rxf7 bxc3 
  2	 -1.83 	32        	0:00.00	bxc3 dxe5 
  1	 -1.83 	16        	0:00.00	bxc3 dxe5 
  0	# 
stockfish developement version find the mate in 6 at depth 20 in less than a second and I see no reason to find mate in 6 in 13 plies.
The interesting thing is time to find the right move and not reported depth and it is known that programs do a lot of pruning.

In this specific case many moves win so I see no important reason to find fast one specific move that only mate faster.

OliverBr
Posts: 655
Joined: Tue Dec 18, 2007 8:38 pm
Location: Munich, Germany
Full name: Dr. Oliver Brausch
Contact:

Re: Tactical positions- some engines get it, some not.

Post by OliverBr » Wed Sep 23, 2020 6:29 am

Uri Blass wrote:
Wed Sep 23, 2020 1:08 am
SF11 is an old version and based on your analysis it reject Ra6 in less than a second at depth 21
21 +3.51 1.87M 0:00.91 Qxc5 Qxc5 Bxc5 a5 Bd4 Rg8 Rc7 Rb8 Be5 Rbf8 f3 Bh4 Ra7 Bd8 Kh2 Bf6 Rxa5 Ra8 Rb5 Bxe5+ Rxe5 Rgb8 Rb5 Rxb5 axb5
It wasn't enough. SF11 actually lost this game by playing Ra6. I wouldn't say SF11 is an old version, many consider it better than newer versions.
Chess Engine OliThink: http://brausch.org/home/chess
OliThink GitHub:https://github.com/olithink

OliverBr
Posts: 655
Joined: Tue Dec 18, 2007 8:38 pm
Location: Munich, Germany
Full name: Dr. Oliver Brausch
Contact:

Re: Tactical endgame position- some engines get it, some not.

Post by OliverBr » Wed Sep 23, 2020 6:35 am

Terje wrote:
Tue Sep 22, 2020 10:56 pm
You don't need to find the best move to win, you just need to find a move that is good enough. Looking for the best move will let you find this mate instantly, but will make you unable to find good enough moves in countless other positions that require you to look deeper :)
Very true. Of course, other moves win, too. And so it doesn't affect the strength of the engine, when this particular move is found later.
The previous puzzle was also cool, Weiss takes a while to realize Rxe3 refutes Ra6, but then immediately gives Rxg7 with a much higher score than the original line had :D
Sometimes the best puzzles come from real games, like this one.
Does Weiss extend every check, no matter the depth? This may be the clue to this position, as here it looks necessary to do so.

Chess Engine OliThink: http://brausch.org/home/chess
OliThink GitHub:https://github.com/olithink

Terje
Posts: 287
Joined: Tue Nov 19, 2019 3:34 am
Location: https://github.com/TerjeKir/weiss
Full name: Terje Kirstihagen

Re: Tactical endgame position- some engines get it, some not.

Post by Terje » Wed Sep 23, 2020 12:50 pm

Weiss extends all checks (actually any time when in check), but that's only half the moves in the line so it takes a while before it plays it out far enough :)

OliverBr
Posts: 655
Joined: Tue Dec 18, 2007 8:38 pm
Location: Munich, Germany
Full name: Dr. Oliver Brausch
Contact:

Re: Tactical endgame position- some engines get it, some not.

Post by OliverBr » Sat Sep 26, 2020 10:37 am

Terje wrote:
Wed Sep 23, 2020 12:50 pm
Weiss extends all checks (actually any time when in check), but that's only half the moves in the line so it takes a while before it plays it out far enough :)
You *really* made some progress with Weiss. Weiss 1.1 is 400 ELO stronger than OliThink, while Weiss 0.5 is 400 ELO weaker. Now testing 0.8, still weaker than OliThink.
Chess Engine OliThink: http://brausch.org/home/chess
OliThink GitHub:https://github.com/olithink

Post Reply