Author: Richard Pijl
Date: 07:34:40 12/15/05
Go up one level in this thread
>Of course the knight can win a move but you still may detect unstoppable pawn if
>the knight need to win 2 moves to stop the pawn or cases that the knight cannot
>win a move thanks to calculation of the distance to the opponent king.
These cases are very very rare, especially in practical play.
> Besides, in most cases the
>>'stopping' capability of the knight is more restricted by pawn and king
>>positions than anything else anyway.
>
>Of course it is clear that distance function cannot detect all the cases
>when pawns are unstoppable but only part of the cases.
>
> finally, you should also consider
>>alternative stopping fields like the second rank, as sometimes it is possible to
>>reach a stopping field other than the promotion square.
>
>of course but I do not think that checking distance to some squares is
>expensive.
Checking just distance is not expensive. But taking pawn and king positions into
account is, as you cannot fully precompute it.
>I do nothing about it and I did not see Movei lose many games because of not
>detecting unstoppable passed pawns.
I don't think it is very important either, at least when you consider total
playing strength.
However, when you reach a knight ending, knight-deflection is a very important
pattern. I think this type of knowledge will help in finding knight deflection
type of combinations faster.
>I suspect that the main advantage is simply being able to prune lines and search
pruning based on this alone is dangerous, as you can never really be sure that
this is 100% right. You'll need search to be able to correct the evaluation when
it goes wrong.
>faster but the advantage is limited and middle game evaluation is more
>important.
of course.
>Rybka that is probably the best program does not detect it by evaluation as the
>following analysis prove(it shows a score of +12 inspite of the fact that even
>without seeing that black promote with check black queen first)
>
>Note that Movei has big problem in this type of positions today and it's search
>is very unstable and it needs depth 10 to see that white is losing.
>
>[D]k6N/8/8/p7/8/6PP/8/7K w - - 0 1
The Baron has -1.02 in evaluation here, without search.
>
>Analysis by Rybka 1.0 Beta 32-bit:
>
>1.Nf7
> +- (12.21) Depth: 3 00:00:00
>
>Another position when Movei is better in finding black's win but rybka is slower
>in finding score above draw.
>
>[D]k6N/8/8/p7/8/6PP/r6R/7K b - - 0 1
Due to the rooks, Baron's static evaluation gives a score of -4.91 (advantage
for white). In search, the first reported line shows the win for black though:
ply time nodes score pv
10(17)& 0:00.22 111865 +3.50 Ra2xh2 Kh1xh2 a5-a4 g3-g4 a4-a3
Nh8-g6 a3-a2 h3-h4 a2-a1Q h4-h5 Qa1-e1
Kh2-h3 Qe1-h1 Kh3-g3 (Ka8-b7) (Kg3-f4)
After forcing Rxh2, Kxh2, the static evaluation is already 0.92, advantage for
black, so just a qsearch should be enough for the Baron to show a black
advantage.
>
>Analysis by Rybka 1.0 Beta 32-bit:
>
>1...Ra1+
> = (0.00) Depth: 3 00:00:00
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1
> = (0.00) Depth: 4 00:00:00
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2
> = (0.00) Depth: 5 00:00:00
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2 Ra2+ 5.Kg1 Ra1+ 6.Kf2
> = (0.00) Depth: 6 00:00:00
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2 Ra2+ 5.Kg1 Ra1+ 6.Kf2
> = (0.00) Depth: 7 00:00:00
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2 Ra2+ 5.Kg1 Ra1+ 6.Kf2
> = (0.00) Depth: 8 00:00:00 1kN
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2 Ra2+ 5.Kg1 Ra1+ 6.Kf2
> = (0.00) Depth: 9 00:00:00 1kN
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2 Ra2+ 5.Kg1 Ra1+ 6.Kf2
> = (0.00) Depth: 10 00:00:00 2kN
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2 Ra2+ 5.Kg1 Ra1+ 6.Kf2
> = (0.00) Depth: 11 00:00:00 3kN
>1...Ra1+ 2.Kg2 Ra2+ 3.Kg1 Ra1+ 4.Kf2 Ra2+ 5.Kg1 Ra1+ 6.Kf2
> = (0.00) Depth: 12 00:00:00 5kN
>1...Rxh2+ 2.Kxh2 a4 3.Nf7 a3 4.Ne5 a2 5.Nc4 a1Q 6.Ne3
> -+ (-5.28) Depth: 12 00:00:00 39kN
>1...Rxh2+ 2.Kxh2 a4 3.Nf7 a3 4.Ne5 a2 5.Nc4 a1Q 6.Ne3
> -+ (-5.28) Depth: 13 00:00:01 46kN
>1...Rxh2+ 2.Kxh2 a4 3.Nf7 a3 4.Ne5 a2 5.Nc4 a1Q 6.Ne3
> -+ (-5.28) Depth: 14 00:00:01 56kN
>1...Rxh2+ 2.Kxh2 a4 3.Nf7 a3 4.Ne5 a2 5.Nc4 a1Q 6.Ne3
> -+ (-5.28) Depth: 15 00:00:02 71kN
>1...Rxh2+ 2.Kxh2 a4 3.Nf7 a3 4.Ne5 a2 5.Nc4 a1Q 6.Ne3
> -+ (-5.28) Depth: 16 00:00:02 90kN
>
>(, 15.12.2005)
>
>Uri
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.