Free gifts from Gray matter

[GoogleCodeExporter]

Here's one example where Gray matter (using MTDf) gives away material
without any need.

post
setboard r1b1k2r/pp1pnppp/4p3/8/2P1P3/1B3NRq/PBQ1KP1P/R7 b kq - 3 17
# Ok, we start with this position
go
1 682 1 39 Qxg3
2 14 1 171 Qh6 Bxg7
3 316 3 1413 Qh5 Bxg7 Qxf3+
4 -118 7 8144 Qh5 Rxg7 Qc5 Rxf7
5 192 25 36231 Qh5 Rg5 Qh6 Rg4 O-O Bxg7
6 -150 163 227734 Qh5 Bxg7 Rg8 c5 Nc6 Bxe6
move h3h5
1 86 0 49 Bxg7 d5 ccxd5 Kd7 Kd3 Qxf3+
2 -2 0 272 Bxg7 d5 ccxd5 Kd7 Kd3 Qxf3+
3 48 4 7880 Bxg7 d5 ccxd5 Kd7 Kd3 Qxf3+
4 -122 59 72978 Bxg7 d5 ccxd5 Kd7 Kd3 Qxf3+
5 90 76 95012 Bxg7 Rg8 Rg5 Qh3 Rg3 Qh5 Rg5 Qh3 Rg3 Qh5
6 -186 563 827329 Bxg7 Rg8 h4 f5 exf5 exf5 Qxf5
usermove b2g7
# it's important to wait quite long here. It's quite possible we have
# to wait until it stored a faulty value in the transposition table
# here. I did not wait until it finished depth 7, but waited let's
# say a minute (I don't know exactly).
1 30 0 40 Qh3 Nd4 Qh5+ Nf3 Qh3 Nd4 Qh5+ Nf3
2 -2 0 432 Qh3 Nd4 Qh5+ Nf3 Qh3 Nd4 Qh5+ Nf3
3 20 6 10258 Qh3 Nd4 Qh5+ Nf3 Qh3 Nd4 Qh5+ Nf3
4 -97 121 149312 Qh3 Nd4 Qh5+ Nf3 Qh3 Nd4 Qh5+ Nf3
5 52 281 373700 Qh3 Nd4 Qh5+ Nf3 Qh3 Nd4 Qh5+ Nf3
move h5d5
# Then here, Gray gives me its queen ...
# While e.g. Rg8 is of course the natural move.
1 -289 1 90 Bxh8 a5 Rg6 Qd6 Nd4 Nxg6
2 -477 6 10305 Bxh8 a5 Rg6 Qd6 Nd4 Nxg6
3 -446 44 81741 Bxh8 a5 Rg6 Qd6 Nd4 Nxg6
4 -610 250 436077 Bxh8 a5 Rg6 Qd6 Nd4 Nxg6
5 -354 470 769655 Bxh8 Qc6 Bd4 Ng6 h4 Qxe4+

Original issue reported on code.google.com by jonne.z...@gmail.com on 29 Nov 2007 at 9:38

[GoogleCodeExporter]

I was thinking ... could this be because we do not have quiescense (where is my
dictionary :o) search? I think that is obligatory. Suppose Gray searches up to 
depth
5. and in the last ply Gray captures our queen. The position is evaluated as 
awesome,
and stored in the table right? But what if, in the next move, we hit Gray's 
queen?
Instead of stopping at maximum depth, we should switch to continue the search 
with
captures only. I think it's as easy as that. Although I would like to verify 
that
*that* is really the problem here.

Original comment by jonne.z...@gmail.com on 1 Dec 2007 at 7:34

[GoogleCodeExporter]

setboard r1b1k2r/pppp1ppp/2n2n2/2b1p3/4P1q1/P1NP1N1P/1PP1QPP1/R1B1KB1R b KQkq - 
0 7

using minimax only GM moves away her queen. using MTDf GM gives away her queen.

Original comment by jonne.z...@gmail.com on 20 Dec 2007 at 5:04

[GoogleCodeExporter]

We've abandoned MTD(f) for MiniMax.  Gray Matter should no longer give away her 
queen.  ;-)

Original comment by brainix on 29 Jan 2008 at 11:42

[GoogleCodeExporter]

Original comment by brainix on 29 Jan 2008 at 11:42

• Changed state: Fixed