Author: Gareth McCaughan
Date: 20:27:07 09/24/01
Go up one level in this thread
Michel Langeveld wrote:
>> [D]2r1r1k1/pp1nbpp1/4pn1p/q3NN1P/P1pP1B2/2P5/1PQ2PP1/R3R1K1 w - - bm Nxg7;
>> id "ECM.1227";
...
> This position isn't found by Crafty either. Crafty considers Nxd7
> for a while but after that switches to Nxe7. In another post
> this position this position is called for sure won by white.
> So yes there must be a hidden win. Deep Fritz says Nxg7 with
> a score of +1.09 afer 1 hour analyzing.
Crafty 18.11 takes about 90 minutes to fail high on Nxg7, and
finally resolves it after a total of about 21 hours (!) with
a score of 1.68. (This is on a 1GHz Athlon, 96+20Mb hash.)
The following was all done using xboard, with a considerably
smaller hash table (12+2.5, I think).
The PV after about 27 hours:
1. Nxg7 Kxg7 2. Bxh6+ Kg8 3. Re3 Nxe5
4. dxe5 Nxh5 5. Qe4 Bf8 6. Bxf8 Kxf8
7. Qh4 Qb6 8. Re2 Qd8 9. Qxh5 Qd3
10. Re3 Qg6 11. Qf3
If you play Nxg7 Kxg7; Bxh6+ Kxh6, and then let it analyse,
it starts off thinking Black is about a pawn ahead; after
10s it sees +0.9 for White, after 16s +1.3, and after 33s
+4.4. (There's another fail-high a little later, resolving
to +5.1 after about 3m20.) One crucial discovery it seems
to make quite slowly (8s to fail high, 40s to see the whole
story) is that after Nxf7+ Kxh5 White can play g4+, with
devastating effect. I conjecture that programs that spot this
faster will do better on the initial position.
Another thing that takes a while is realising how much
trouble it's in after 3.Re3. For the first 90ish seconds
it thinks Black is ahead. More specifically: after 3.Re3 Nxe5,
it takes about 35s to realise that dxe5 is better than Rxe5.
This in turn seems to be because it's slow to see the strength
of Qe4 after Nxh5. (It takes about 4s before it realises that
it's positive for White.)
Blah. All this tree-climbing is rather uninspiring.
Still, there might be some illumination in seeing how
other programs do in those positions along the way
where Crafty has to think for a while. Specifically,
what do other programs make of this position (after
5.Qe4)? Crafty needs 9 more ply to regard it as positive
for White.
[D]2r1r1k1/pp2bp2/4p2B/q3P2n/P1p1Q3/2P1R3/1P3PP1/R5K1 b - -
--
g
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.