Author: Colin Frayn
Date: 08:36:59 03/01/00
Go up one level in this thread
>>>Yesterday I saw an interesting endgame position on ICC.
>>>[D]8/8/6k1/3p4/3P4/2P2PKp/pB4b1/8 b - -
>>>
>This position is a simple win for black.
>For example:
>1... Kf5
>2.Ba1 Bf1
>3.Bb2 Be2
>4.Ba1 h2
>5.Kxh2 Kf4
>5.Kg2 Bxf3
>6.Kf2 Ke4
>7.Ke1 Kd3
<numbering problem - this is 15 ply i.e. 8 moves>
>and black wins the white bishop and the game.
>White may defend in a different way, but without any hope.
ColChess (Development version) gives the following analysis to 16 plies plus
quiescence extensions. (note that score given is the improvement from the
current positional evaluation)
Current Position
----------------
. . . . . . . .
. . . . . . . . pP=Pawn
. . . . . . k . rR=Rook
. . . p . . . . nN=Knight lower=Black
. . . P . . . . bB=Bishop CAPS=White
. . P . . P K p qQ=Queen
p B . . . . b . kK=King
. . . . . . . .
Thinking [15 Moves]
Current Position=1.68
{L10} :::::::. G6 - F5 (0.03) Nodes=203232 Eval=73604 [4.25s]
{L11} :::::::. G2 - F1 (0.18) Nodes=1336908 Eval=400921 [16.16s]
{L12} :::::::. G2 - F1 (0.18) Nodes=671756 Eval=235773 [22.74s]
{L13} :::::::. G2 - F1 (0.18) Nodes=1364824 Eval=386905 [33.36s]
{L14} :::::::. G2 - F1 (0.18) Nodes=1802620 Eval=594777 [101s]
{L15} :::::::. G2 - F1 (0.18) Nodes=4650125 Eval=1127314 [169s]
{L16} :::::::. G2 - F1 (0.18) Nodes=21361910 Eval=7349591 [711s]
11 Minutes, 51 Seconds
Nodes=31595297 Eval=10238058
Table Hits=2066996 [279620 Entries]
PV : g2f1 b2a1 g6g5 a1b2 g5f6 b2a1 f6e7 a1b2 <HT>
Best = G2 - F1
------------------------------------
...so there's definitely no substantial material gain for either side within 16
ply. ColChess is immune to the problems of null move pruning (as it doesn't use
it). It doesn't use futility pruning or razoring either, so if it says it's not
won by 16 ply then it's correct.
Of course, after the moves you gave, the position is one of those which a human
can spot is easily won, but a computer can't. It would require a lot more ply
of searching to find that. Or a clever pawn&bishop endgame function. Hmmm....
I think that the first person who can write a function
obvious_win(position,side) and implement it in a fast and efficient way will get
as lot of advantages out of it, and will start solving some of the more
difficult endgame problems in ridiculously short times.
If your initial PV is correct then the position after 4. Ba1 is definitely won -
ColChess spots h2 in 14 ply, and the score increases by 15 ply. I'll check 16
ply this evening;
Current Position
----------------
. . . . . . . .
. . . . . . . . pP=Pawn
. . . . . . . . rR=Rook
. . . p . k . . nN=Knight lower=Black
. . . P . . . . bB=Bishop CAPS=White
. . P . . P K p qQ=Queen
p . . . b . . . kK=King
B . . . . . . .
Thinking [12 Moves]
Current Position=1.81
{L10} :::::: E2 - F1 (-0.10) Nodes=223128 Eval=88205 [5.25s]
{L11} :::::: E2 - F1 (-0.10) Nodes=412139 Eval=180203 [9.87s]
{L12} :::::: E2 - F1 (-0.10) Nodes=797908 Eval=317812 [17.82s]
{L13} :::::: E2 - F1 (-0.10) Nodes=1902180 Eval=699556 [35.89s]
{L14} :::::: H3 - H2 (0.14) Nodes=28985880 Eval=10663013 [827s]
{L15} :::::: H3 - H2 (0.42) Nodes=48003080 Eval=17075509 [2145s]
Cheers,
Col
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.