How to get chess program to solve KBN mate?

Paul Pedriana

unread,

Sep 17, 1997, 3:00:00 AM9/17/97

to

I am trying to get my chess program to solve a mate
given that white has a King, Bishop, and Knight,
while Black has only a King. It can't solve it, no
matter how hard it tries. It is ChessMaster 5000

Can any chess program solve this?

Paul

Robert Hyatt

unread,

Sep 17, 1997, 3:00:00 AM9/17/97

to

Paul Pedriana (pped...@maxis.com) wrote:
: I am trying to get my chess program to solve a mate

Most can. the key piece of information needed is "what is the
right corner?" If a program understands this, all it needs to
know is to drive the king to the edge of the board, and then
to the right corner, and a decent search will find the mate once
you get the king to the right corner...

endgame databases solve this quite nicely, of course...

Chuck Cadman

unread,

Sep 17, 1997, 3:00:00 AM9/17/97

to

Your saying that a program without an endgame database could solve this
ending based on knowing which corner to drive the king into? How long does
it take?

Chuck Cadman

Robert Hyatt

unread,

Sep 17, 1997, 3:00:00 AM9/17/97

to

Chuck Cadman (cdca...@midway.uchicago.edu) wrote:
: Your saying that a program without an endgame database could solve this

: ending based on knowing which corner to drive the king into? How long does
: it take?

: Chuck Cadman

Not a long time. What you need is something to let it know (a) drive king
to the edge of the board and (b) then drive it to the right corner. It does
take some searching, but if you give a program maybe 30 seconds per move it
should be easy. If you give a program 10 seconds it should still be able
to do it, as there are direct search solutions to the mini-problems it
has to solve. It won't find the mate in 30 from a bad-case position of
course, but it will be able to mate the opponent in a reasonable number of
moves. It should see the mate at around 8-10 moves left, maybe, but it
will be heading for the right goal all the time.

You might try crafty without KBN vs K databases, to see how it does. It
should solve this given a reasonable amount of time. What is reasonable
depends on your machine. But it should pull it off...

David John Blackman

unread,

Sep 18, 1997, 3:00:00 AM9/18/97

to

"Chuck Cadman" <cdca...@midway.uchicago.edu> writes:

> Your saying that a program without an endgame database could solve this
>ending based on knowing which corner to drive the king into? How long does
>it take?

>Chuck Cadman

Andrew Tridgell and i added this to (one of the versions of) KnightCap
about a month ago. It took us less than an hour including a bit of testing.
Just do a search of 8 ply or so knowing the losing king belongs as close as
possible to a corner the bishop can attack, and the two kings should be
close together. The longest line of resistance from the worst position is
supposed to be about 30 moves.

Of course this ending hardly ever happens (about once per career
according to Capablanca), but i suspect a lot of programs can do it just
for the bragging rights.

brucemo

unread,

Sep 18, 1997, 3:00:00 AM9/18/97

to

David John Blackman wrote:

> Andrew Tridgell and i added this to (one of the versions of) KnightCap
> about a month ago. It took us less than an hour including a bit of testing.
> Just do a search of 8 ply or so knowing the losing king belongs as close as
> possible to a corner the bishop can attack, and the two kings should be
> close together. The longest line of resistance from the worst position is
> supposed to be about 30 moves.

34 according to Fine, but he is wrong. The correct answer is 33.

> Of course this ending hardly ever happens (about once per career
> according to Capablanca), but i suspect a lot of programs can do it just
> for the bragging rights.

It happened at the WMCCC Jakarta last year.

A good reason to implement KBN vs K is to avoid, "I bought this program and
it can't even mate with a bishop and a knight" discussions, which happen lots
more often than once a career :-)

bruce