Re: computer-go: Go4++: Passing in tsume searches.

["Michael Reiss" <mick@xxxxxxxxxxxxxxxxx>]

> >Are there any tsume problems where the defender *must* pass *more than
> >once* (not necessarily consecutively) in order to prove it's alive or
> >seki?
> >
> >Are there any tsume problems where the defender *must* pass but the pass
> >move is not the defenders last move before the search terminates?
>
> Look at the following problem:
>
>  6 * * * . . . . .
>  5 o o * * * . . .
>  4 . o o o * * . .
>  3 o * o o * * . .
>  2 * * . o * * . .
>  1 . o o o * * . .
>    A B C D E F G H
>
> Assume o has just captured * on A4 and * is to move next.
> The optimal play has lots of passes:
>
> * passes,    o passes,
> * A4,        o passes,
> * C2,        o A3,
> * ko-threat, o ko-answer,
> * A4,        o passes,
> * A1
>
> By adding outside liberties for o one can increase the number of
> passes of o arbitrarily (and by that increase the number of necessary
> ko-threats if more than 3 outside liberties for o are added).

But under my scheme the problem looks very different. If it's doing
"o wins all ko's" then * is not allowed to capture o at A4 because it
starts a ko. In this case the "defending" group is simply "ALIVE".

A possible sequence is:
* A1(A4 not allowed)    o C2 - two eyes detected - o ALIVE.

When I do the search with "* wins all ko's" then * is free to play
A4 any time and o is not allowed to capture back because it starts
a ko. In this case the defending group is simply "DEAD".

A possible sequence is:
* A4 (Allowed)   oA1 (A3 not allowed)
* C2 - o is DEAD

Under my system, when there is a difference between the result
depending on who wins ko's then the final answer is simply "KO".

I know my system is crude because it takes no account of the
number of ko threats required, so in the following situation
my tsume search simply states that the status of o is "KO".
Luckily Go4++ will detect that o has only got space for one
eye and so it knows that it will die.

>  5 . . . . . . . .
>  4 . . . . . . . .
>  3 * * * * * . . .
>  2 o o o o * * . .
>  1 o . o . o * . .
>    A B C D E F G H

Mick.