Stupid for human but not for all engine

[brennus]

Hello,

5bk1/p3p1p1/p3P1P1/p7/p7/p7/P7/7K w - - 0 1

48/47 01:00 292.874.450 4.876 +M18 Rg2 Rh8 Rf3 Rg8 Re4 Rh8 Rd5 Rg8 Rc6 Rh8 Rb7 Rg8 Rxa7 Rh8 Rxa6 Rg8 Rxa5 Rh8 Rxa4 Rg8 Rxa3 Rh8 Rb3 Rg8 a4 Rh8 a5 Rg8 a6 Rh8 a7 Rg8 a8D Rh8 Dxf8+

Some of modern engine can't find the way to solve this and i feel it is strange

With help of HashTable it should be immédiate because only 2 legal position for black.

It it the limit of deep search ? (here 48hply) many positions should be the same and be found in HT (at least before the fist take of pawn)

Some engine found it in less than the second, this one in one minute some other never found this stupid position.

is it a test for HT management or for deep seek limit ?

[Kurt Utzinger]

Hello,

[d]5bk1/p3p1p1/p3P1P1/p7/p7/p7/P7/7K w - - 0 1

48/47 01:00 292.874.450 4.876 +M18 Rg2 Rh8 Rf3 Rg8 Re4 Rh8 Rd5 Rg8 Rc6 Rh8 Rb7 Rg8 Rxa7 Rh8 Rxa6 Rg8 Rxa5 Rh8 Rxa4 Rg8 Rxa3 Rh8 Rb3 Rg8 a4 Rh8 a5 Rg8 a6 Rh8 a7 Rg8 a8D Rh8 Dxf8+

Some of modern engine can't find the way to solve this and i feel it is strange

With help of HashTable it should be immédiate because only 2 legal position for black.

It it the limit of deep search ? (here 48hply) many positions should be the same and be found in HT (at least before the fist take of pawn)

Some engine found it in less than the second, this one in one minute some other never found this stupid position.

is it a test for HT management or for deep seek limit ?

[zamar]

Stockfish (current devel snapshot) sees this in less than a second. However it's funny that because of aggressive reductions in sidelines it requires depth of 77 plies to see the solution!

2 -7.15 00:00 5 Kg2 Kh8
3 -6.95 00:00 22 Kg2 Kh8 Kf3
4 -7.15 00:00 44 Kg2 Kh8 Kf3 Kg8
5 +0.00 00:00 74 Kg2 Kh8 Kf3 Kg8 Kg2
6 +0.00 00:00 111 Kg2 Kh8 Kf3 Kg8 Kg2
7 +0.00 00:00 152 Kg2 Kh8 Kf3 Kg8 Kg2
...

76 +0.00 00:00 175076 Kg2 Kh8 Kf3 Kg8 Kg2
77 > #20 00:00 209517 Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8
Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8 Kb2
Kg8 a4 Kh8 a5 Kg8 a6
78 #18 00:00 270577 Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8
Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8 Kb2
Kg8 a4 Kh8 a5 Kg8 a6 Kh8 a7 Kg8 a8=Q Kh8 Qxf8#
79 #18 00:00 292200 Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8
Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8 Kb2
Kg8 a4 Kh8 a5 Kg8 a6 Kh8 a7 Kg8 a8=Q Kh8 Qxf8#
80 #18 00:00 312408 Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8
Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8 Kb2

[Aaron Becker]

Nice position. I tested it on Daydreamer, which hit its search depth limit in under a second without finding anything. After I bumped it up (from 64 to 128), it finds the solution very quickly. Clearly I was silly to think that 64 was a very conservative upper limit for search depth.

info multipv 1 depth 2 score cp -641 time 1 nodes 5 qnodes 4 pvnodes 1 nps 2000 tbhits 0 pv h1g1 g8h8
info string sanpv Kg1 Kh8 Kf2
info multipv 1 depth 3 score cp -613 time 1 nodes 17 qnodes 12 pvnodes 3 nps 8000 tbhits 0 pv h1g1 g8h8 g1f2
info string sanpv Kg1 Kh8 Kf2 Kg8
info multipv 1 depth 4 score cp -640 time 1 nodes 33 qnodes 19 pvnodes 6 nps 16000 tbhits 0 pv h1g1 g8h8 g1f2 h8g8
...
info multipv 1 depth 64 score cp 0 time 126 nodes 247046 qnodes 52254 pvnodes 246 nps 1945000 tbhits 0 pv h1g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8 f2g1 g8h8 g1f2 h8g8
info multipv 1 depth 65 score mate 21 time 210 nodes 445602 qnodes 79961 pvnodes 287 nps 2111000 tbhits 0 pv h1g1 g8h8 g1f2 h8g8 f2e3 g8h8 e3e4 h8g8 e4d5 g8h8 d5c6 h8g8 c6b7 g8h8 b7a6 h8g8 a6a5 a7a6 a5a4 a6a5 a4a3 a5a4 a3a4 g8h8 a4a5 h8g8 a5a6 g8h8 a2a4 h8g8 a4a5 g8h8 a6b5 h8g8 a5a6 g8h8 a6a7 h8g8 a7a8q g8h8 a8f8

[Edsel Apostol]

TL finds it in an instant at depth 40.

Code: Select all

FEN: 5bk1/p3p1p1/p3P1P1/p7/p7/p7/P7/7K w - - 0 1 

T20091209:
  8/4	00:00	         143	0	 0.00	Kh1g2 Kg8h8 Kg2h1 Kh8g8
  9/4	00:00	         165	0	 0.00	Kh1g2 Kg8h8 Kg2h1 Kh8g8
 ..........................
 39/33	00:00	      27.483	0	 0.00	Kh1g2 Kg8h8 Kg2h1 Kh8g8
 40/43	00:00	      44.727	0	+M18	Kh1g2 Kg8h8 Kg2f3 Kh8g8 Kf3e4 Kg8h8 Ke4d5 Kh8g8 Kd5c6 Kg8h8 Kc6b7 Kh8g8 Kb7xa7 Kg8h8 Ka7xa6 Kh8g8 Ka6xa5 Kg8h8 Ka5xa4 Kh8g8 Ka4xa3 Kg8h8 Ka3b2 Kh8g8 a2a4 Kg8h8 a4a5 Kh8g8 a5a6 Kg8h8 a6a7 Kh8g8 a7a8Q Kg8h8 Qa8xf8+

[JVMerlino]

Even my relatively dumb engine, Myrddin, can find the answer in less than two seconds, but only if I increased its max search depth. The release version has a max depth of 50, but it needed to get to depth 53 to announce mate.

Code: Select all

50/31     0     20     80214  h1g2 g8h8 g2f3 h8g8 f3g2
51/31     0     21     89858  h1g2 g8h8 g2f3 h8g8 f3g2
52/31     0     23    102018  h1g2 g8h8 g2f3 h8g8 f3g2
53/39    40     29    147234  h1g2!
53/39   340     31    158391  h1g2!
53/51 32732    167   1489739  h1g2 g8h8 g2f3 h8g8 f3e4 g8h8 e4d5 h8g8 d5c6 g8h8
c6b7 h8g8 b7a7 g8h8 a7a6 h8g8 a6a5 g8h8 a5a4 h8g8 a4a3 g8h8 a3b4 h8g8 a2a4 g8h8
a4a5 h8g8 a5a6 g8h8 a6a7 h8g8 a7a8Q g8h8 a8f8
54/51 32732    232   1988874  h1g2 g8h8 g2f3 h8g8 f3e4 g8h8 e4d5 h8g8 d5c6 g8h8
c6b7 h8g8 b7a7 g8h8 a7a6 h8g8 a6a5 g8h8 a5a4 h8g8 a4a3 g8h8 a3b4 h8g8 a2a4 g8h8
a4a5 h8g8 a5a6 g8h8 a6a7 h8g8 a7a8Q g8h8 a8f8

jm

[Uri Blass]

Hello,

5bk1/p3p1p1/p3P1P1/p7/p7/p7/P7/7K w - - 0 1

48/47 01:00 292.874.450 4.876 +M18 Rg2 Rh8 Rf3 Rg8 Re4 Rh8 Rd5 Rg8 Rc6 Rh8 Rb7 Rg8 Rxa7 Rh8 Rxa6 Rg8 Rxa5 Rh8 Rxa4 Rg8 Rxa3 Rh8 Rb3 Rg8 a4 Rh8 a5 Rg8 a6 Rh8 a7 Rg8 a8D Rh8 Dxf8+

Some of modern engine can't find the way to solve this and i feel it is strange

With help of HashTable it should be immédiate because only 2 legal position for black.

It it the limit of deep search ? (here 48hply) many positions should be the same and be found in HT (at least before the fist take of pawn)

Some engine found it in less than the second, this one in one minute some other never found this stupid position.

is it a test for HT management or for deep seek limit ?

Your position is illegal

I changed it to a legal position that rybka cannot solve

see
http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=14109

[Sven]

Rybka 2.2n2 mp 32-bit on my old PIV under WinBoard: 0.13 seconds

Output in analysis window:

Code: Select all

...
 36	+99.65	22923	0:00.24	Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8 Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8
 35	+99.65	18994	0:00.20	Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8 Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8
 34	+99.65	15505	0:00.19	Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8 Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8
 33	+99.65	12481	0:00.17	Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8 Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8
 32	+99.65	9886	0:00.14	Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8 Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8
 31	+99.65	7735	0:00.14	Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8 Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8
 30	+99.65	5975	0:00.13	Kg2 Kh8 Kf3 Kg8 Ke4 Kh8 Kd5 Kg8 Kc6 Kh8 Kb7 Kg8 Kxa7 Kh8 Kxa6 Kg8 Kxa5 Kh8 Kxa4 Kg8 Kxa3 Kh8
 29	0.00	3917	0:00.11	Kg2 Kh8 Kf3 Kg8 Kg2 Kh8 Kf3 Kg8 Kg2 Kh8 Kf3 Kg8 Kg2 Kh8
 28	0.00	3186	0:00.08	Kg2 Kh8 Kf3 Kg8 Kg2 Kh8 Kf3 Kg8 Kg2 Kh8 Kf3 Kg8 Kg2 Kh8
 27	0.00	2695	0:00.08	Kg2 Kh8 Kf3 Kg8 Kg2 Kh8 Kf3 Kg8 Kg2 Kh8 Kf3 Kg8 Kg2 Kh8
...

Sven

[Jouni]

Very familiar position based on my mate in 17 position :

[d]5bk1/p3p1p1/p3P1P1/p7/p7/8/P7/5K2 w - -

Jouni

[brennus]

This original test come from Pierre Nolot i think...