EGTB - Is This Normal?

[rcmaddox]

In this position, white mates in 16 moves.

[d]8/5R2/5R2/2k5/6K1/r7/5P2/8 w - - 5 80

However, using only 5 man Syzygy TB's, Houdini 4 played Rc6+, jettisoning a rook! Houdini still wins, but goes the long way around the barn to do it. So, not having the complete 6 man set, it saw a win with 5 pieces and went for it?

[syzygy]

So, not having the complete 6 man set, it saw a win with 5 pieces and went for it?

Correct. Happens all the time, also with Nalimov.

[peter]

, also with Nalimov.

Hi!

Nalimov's 5:

Code: Select all

8/5R2/5R2/2k5/6K1/r7/5P2/8 w - -

Engine: Houdini 4 Pro x64 (8192 MB)
von Robert Houdart
 ...
9/19   0:00   +M39     1.Tc6+ Kb5 2.Tb7+ Kxc6 (29.388) 1632 TB:25 
...
15/19  0:00   +M36     1.Tc6+ Kb5 2.Tc5+ Kxc5 (47.682) 1083 TB:735 
... 
20/23  0:00   +M35     1.Tc6+ Kb5 2.f4 Ta8 3.Tc5+ Kxc5 (1.388.549) 8364 
                       TB:50.817 
...
22/27  0:00   +M34     1.Ta7 Td3 2.Tc7+ Kb5 3.Tb6+ Ka5 
                       4.Tb8 Ka6 5.f4 Td1 6.Ta7+ Kxa7 (5.247.884) 12831 
                       TB:174.752 
...
26/33  0:02   +M32     1.Tc7+ Kb5 2.Tb6+ Kxb6 (43.813.849) 16440 
                       TB:1.058.764 
27/39  0:04   +M31     1.Tc7+ Kb5 2.Tf8 Ta4+ 3.f4 Kb6 4.Tc2 Kb7 
                       5.Tb8+ Kxb8 (71.123.247) 16679 
                       TB:1.684.553 
28/39  0:06   +M21     1.Tc7+ Kb5 2.Tf5+ Kb4 3.Tf8 Ta5 
                       4.Tb8+ Ka3 5.Tcb7 Ta4+ 6.Kf3 Ka2 
                       7.Tb2+ Ka1 8.Tb1+ Ka2 9.T8b2+ Ka3 
                       10.Ta1+ Kxb2 (112.816.252) 16479 
                       TB:2.586.717 
... 
35/41  1:11   +M20     1.Tc7+ Kd5 2.Tcc6 Ta2 3.Kg3 Ta7 
                       4.Tfd6+ Ke4 5.f3+ Ke3 6.Tc4 Te7 
                       7.Td1 Ke2 8.Te4+ Txe4 (1.179.065.664) 16485 
                       TB:20.622.013 
36/41  1:57   +M19     1.Tc7+ Kd5 2.Tf5+ Kd4 3.Tf8 Ke5 
                       4.Td7 Ta2 5.Te8+ Kf6 6.Td6+ Kf7 
                       7.Ted8 Txf2 (1.948.762.946) 16560 
                       TB:32.520.018 
...
39/41  6:06   +M17     1.Tc7+ Kd5 2.Tf5+ Kd4 3.Tf8 Ke5 
                       4.Td8 Ta4+ 5.Kg5 Ke4 6.Td2 Ta1 
                       7.Te7+ Kf3 8.Te3+ Kg2 9.f4+ Kf1 
                       10.Kg4 Kg1 11.Ted3 Tc1 12.Kf3 Tf1+ 
                       13.Kg3 Te1 14.Td1 (6.202.619.484) 16914 
                       TB:83.762.440 
...
42/43  14:20  +M16     1.Tc7+ Kd5 2.Tf5+ Kd6 3.Th7 Ta4+ 
                       4.f4 Td4 5.Th6+ Kc7 6.Tf7+ Td7 
                       7.Txd7+ Kxd7 (12.213.056.122) 14187 
                       TB:142.276.543

[Vinvin]

... So, not having the complete 6 man set, it saw a win with 5 pieces and went for it?

Sure ! The engines play the shortest win it "see".

[syzygy]

, also with Nalimov.

Hi!

Nalimov's 5:

It is true that an engine probing WDL tables as opposed to DTM tables during search "suffers" more from it, but as you see also using Nalimov there is at first a preference for the quick conversion into a known win.

In the example position, given enough time H4+syzygy (5-piece) will probably find a way to mate while keeping 6 pieces on the board. On second thoughts, probably not: black will prevent getting mated (in the search) by forcing white to take its rook....