Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: number of pawn positions in chess...

Author: Ricardo Gibert

Date: 19:20:16 01/13/00

Go up one level in this thread


On January 13, 2000 at 21:39:49, Ricardo Gibert wrote:

>On January 12, 2000 at 18:34:49, Dan Ellwein wrote:
>
>>Hi
>>
>>Just wanted to bounce this off of the group and see if this is an accurate
>>representation of how many (non redundant) pawn positions there are in chess...
>>
>>
>>(0,8)(1,7)(2,6)(3,5)(4,4) 8P 48x47x46x45x44x43x42x41  x5 = _______
>>(0,7)(1,6)(2,5)(3,4)      7P 48X47X46X45X44X43X42     x4 = _______
>>(0,6)(1,5)(2,4)(3,3)      6p 48X47X46X45X44X43        x4 = _______
>>(0,5)(1,4)(2,3)           5P 48x47x46x45x44           x3 = _______
>>(0,4)(1,3)(2,2)           4P 48x47x46x45              x3 = _______
>>(0,3)(1,2)                3P 48x47x46                 x2 = _______
>>(0,2)(1,1)                2P 48x47                    x2 = _______
>>(0,1)                     1P 48                       x1 = _______
>>
>>
>>Number of non redundant pawn postions in chess  --  TOTAL: _______
>>
>>haven't done the math on it yet, but it looks like about 75 trillion...
>>
>>thanks...
>>
>>PilgrimDan
>
>I wrote a short little program that calculates the number of pawn positions to
>be 8e10. The vast majority of positions represented by this number are illegal.
>so this is an upper bound. However, I'm a C newbie so I probably made an error.
>
>In any case, there is a quite simple argument that shows the upper bound to be
>less than 1.8e19: Consider a 48 bit bitmap giving the locations of all the pawns
>on the board. Since at most 16 of these bits can be set, we can use a 16 bit
>bitmap to give color. That's 64 bits total or 1.8e19 pawn positions.
>
>Of course, we are overcounting like crazy. The 48 bit location bitmap includes
>positions with 17 to 48 pawns! Also, the 16 bit color bitmap includes those
>positions with more than 8 white or 8 black pawns.
>
>Taking all this into account I calculated the 8e10 result. With a little
>friendly cajoling, I may be persuaded to explain how I calculated this result.
>In the meantime, I will double check my result.

8e10 is an error. I found and corrected the overflow problem I feared and came
up with 6.1e12. I'll triple check now...



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.