|
Comments
After I learnt that position 91 still resisted to present day programs and hardware, I was a little more cautious to expect a solution here. Nevertheless, Heiner Marxen (HM) wrote a special purpose program to solve mates, and way back in 1997, his program "Chest" came up with a mate in 12 on a Solaris machine with 70MB hash. At that time it needed 36 hours to get to the solution. Sadly enough, the main line was not given.
So Fritz6 got a first shot again at this hard dish. After the initial and "obvious" 1.Nf3+, the program switches to 1.Qc7 after 2:46 minutes, to find 1.Nc3 after "only" 08:51 minutes. The evaluation goes up to +5.69 @ ply 15/15 (1:51:04 & 10061239 kN).
Shredder7 finds 1.Nc3 after only 10 seconds, and keeps this as best move throughout. The initial evaluation is +3.49. The first time it sees that 1.Nc3 is going to mate, is after only 90 seconds: mate in 18. It's only at ply 29 that a mate in 12 is found.
After a busy period, in which testing shifted to the background, I decided to display two times (see background info form the computer test start page): the first time the correct move was found (and kept) and the last time (mostly the "mate announcement" time).
So Hiarcs was the first program to be kept record of this way: first time Nc3 was displayed: 1min53s; +4.23, 23850 kn, ply ?/?; penultimate time: 0:31:32, +8.95, 380430 kn, ply 12/12. with analysis: 1.Nc3+ Kxf1 2.Nf3+ Kxg2 3.Rxd1 Nf6+ 4.Kxf7 Nc6 5.Qxc7 e5 6.d3 Bb4 7.Rg1. The last displayed info was at 12/31: +9.20: 1.Nc3 Kxf1 (01:52:36, 1322467 kn).
This is obviously one of my favourite positions. It's a tough one and you can easily leave the program running overnight to get a decent solution. So the next guinea pig was Chess Tiger 14, also as an engine under the Chessbase GUI (I hope to rerun these tests in the future with the engines under the Fritz GUI). The display gave this from ply 11 onwards:
ply 11/.. 1.Nf3+ Kxf2 2.Nc3 Nf6+ 3.Kxf Qxc2 4.d3 Nc6 5.Qxc7 d6 6.Bg5 Be6+ 7.Kxe6 Nc8+ 8.Qxd8 Rxd8 +4.44 35s 17835kn
and here, surprisingly, no fail low was displayed, but immediately the improvement:
ply 12/.. 1.Nc3 +5.34 03:07 95389kn
ply 12/.. 1.Nc3 Qxc2 2.d3 Kxf1 3.Qxc7 Nf6+ 4.Kxf7 d6 5.Nf3+ Kxg2 6.Qxc8 Nbd7 7.Rg1+ Kxf2 8.Ne4 +6.32 03:22 102779kn
ply 13/.. 1.Nc3 Qxc2 2.d3 Kxf1 3.Qxc7 Nf6+ 4.Kxf7 d6 5.Nf3+ Kxg2 6.Rg1+ Kxf2 7.Ne4+ Nxe4 8.Qxc2 Bh3 9.dxe4 +6.64 04:38 143121kn
ply 14/.. 1.Nc3 Kxf1 2.Nf3 Kxg2 3.Rxd1 Nf6+ 4.Kxf7 Nc6 5.Qxc7 e5 6.d4 Bb4 7.Ne1+ Kxh2 8.Nf3+ Kg2 9.Rg1 +7.36 17:22 552022kn
When playing through this last line, the mate is not far away anymore, but is the preceding line correct?
ply 15/.. 1.Nc3 +8.26 00:42:43 1314780kn
and finally, after a night calculating:
ply 15 or 16 1.Nc3 Kxf1 2.Nf3+ Kxg2 3.Rxd1 Nf6+ 4.Kxf7 Nc6 5.Rg1+ Kxf2 6.Nd1+ Kxe2 7.Nd4+ Nxd4 8.Qxc7 Ne4 M13 04:23:16 8957635kn
I write 15 or 16, because when I checked the program in the morning, it still displayed depth 16, and was still checking b1-c3 (1st move out of 26) at a speed of 563 kn/s. So I cannot say for sure whether the displayed mate was found at ply 15 or 16 - a personal guess would be 15, the previous display being a fail high.
Fritz6 does not find the solution in the short time given (11 minutes) - I think it was detecting that 1.Nf3 was not the best move, as on ply 10 (+4.59) and ply 11 (+4.28), the evaluations were higher. More testing is for later - total number of nodes: 39.282 kns (average search speed: 307 knps).
|
