Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Explanation of EGTB Statistics

Author: Dieter Buerssner

Date: 14:09:27 02/28/04

Go up one level in this thread


On February 28, 2004 at 16:36:32, David H. McClain wrote:

>For the 5 piece knnkn.nbb.emd

I guess what follows is not only statistics for knnkn.nbb.emd, but rather
statistics for both, knnkn.nbb.emd (the later part) and knnkn.nbw.emd (the
former part of the table).

>wtm: Lost in   0:            56
>wtm: Draws:            44082657
>wtm: Mate in   7:            38
>wtm: Mate in   6:           757
>wtm: Mate in   5:          3887
>wtm: Mate in   4:          5955
>wtm: Mate in   3:          9382
>wtm: Mate in   2:         10535
>wtm: Mate in   1:          4973
>btm: Lost in   0:           964
>btm: Lost in   1:          1007
>btm: Lost in   2:          1404
>btm: Lost in   3:           726
>btm: Lost in   4:           796
>btm: Lost in   5:           177
>btm: Lost in   6:            18
>btm: Draws:            48090675
>btm: Mate in   1:           293
>
>Could someone interpret these figures to a layman?

wtm means white to move, btm black to move. However, these are perhaps not the
best abbrevetions. Actually wtm means "left side" to move (left side in knnkn,
so knn side) and btm means the side with the one knight and the king has the
right to move.

For the side with the 2 knights, it turns out, that there are
56 + 44082657 + 38 + 757 + 3887 + 5955 + 9382  + 10535 + 4973 "different"
possible posisions. Obviously, most of those are draws. Very few are "lost in
0". This just means, that the NN side is mated already in those positions. Mate
in 1 means just, can force mate in 1. Similar to the others. "btm: Lost in 6:
..." means, after my best move, opponent can force a mate in 6.

Some elobaration on "different". There are no pawns in this endgame. So we can
do many symmetry operations on the board, without changing the prinicpal
position. Basically, you can force by mirroring operations on the board (for
example) the king with the 2 knights to be always inside the triangle
a1-d1-d4-a1. There are even few more restrictions (for example, when the K is on
the long diagonal a1-h8, you can apply more syymetry operations, and force the
opponent K to not be above that diagonal).

So, only positions, that are not transferrable to another already counted
position by any symmetry operation are counted. (Most probably some positions
are counted more than once - namely those, where both kings are on the same long
diagonal.)

You see, that both TBs (nbb and nbw) have something like 50 million "different"
positions. We can try to estimate this by some math. Without symmetry
considerations, and considerations about illegal positions (because the side not
to move would be in check) we could estimate, that there are

64*63*62*61*60/2

different positions. The first piece we place on the board has 64 possible
squares, the second has 63 left, ...

The divisor of 2, is because the 2 knights of the same color are
indistinguishable. First knight on a1 and second knight on a2 is the same as
first knight on a1 and second knight on a2. If we force the "white" (=side of
the two knights) K to the above triangle, this will be

10*63*62*61*60/2

We can estimate a little bit better, by enumerating all legal KK positions with
the restrictions noted above. It turns out, that there are not 10*63 such
positions, but rather only 462 such positions. So the last estimate will be

462*62*61*60/2 =  52418520

A bit over 50 millions. The ones, that are missing in the above table above are
the ones, that are because side not to move is in check by an opponent knight
(and therfore illegal).

Regards,
Dieter



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.