RE: computer-go: Live or Die

[Jean-Pierre Vesinet <jvt@xxxxxxxxxx>]

> Benson's algorithm is mathematically elegant, but it does not help you
> that
> much in practice. If you run it on the test data from my paper you will
> see
> many obviously alive areas that are not Benson-alive.
> 
	Certainly. Bensons's algorithm is mainly the "rock bottom" for other
algorithms and theories. 

	BTW, I recall a diagram in your paper (the first or second diagram)
where two similar positions where presented in a left-right symmetry. The
only difference between the two was a single stone. One of the position was
supposed to be safe, the other unsettled. Pardon me if I'm wrong but it
seemed to me that the "safe" one was actually a ko. I'll have to check this,
and I'll post my "solution" tomorrow.

> A region is a connected set of points on the Go board
> which is surrounded by blocks of the same color.
> A block is called enclosing block of a region if it
> has at least one adjacent point that belongs to the region, and
> at least one adjacent point that does not belong to the region.
> The color of enclosing blocks is called the defender,
> the other color is called the attacker.
> 
> The interior of a region is the subset of points not adjacent
> to an enclosing block.
> There may be both attacker and defender stones in the interior.
> 
	Well, I don't want to criticise these definitions; after all,
everybody knows what "surrounded" means in Go. :). 

	When I read your paper last year I was delighted to find ideas (that
were new to me) about L&D recognition. Since then I have been trying to
improve the definitions, scope and extend your ideas. I hope you don't mind
a bit of (constructive) criticism of your paper. I'll post some remarks
tomorrow (when I find my notes :).

	IMHO there are some loopholes in the way the criteria are worded. My
concern is also about the completeness of the criteria defining what is
known as "static" life (called "unconditional life"  in your paper). The
word "unconditional" has been used to mean (too many) different things in Go
literature.

> The best currently existing static rules are probably those in Thomas
> Wolf's program. He has written several papers about his program. However,
> there is little detail information about the rules he uses. His focus
> seems
> to be on small areas occurring in Life-and-Death problems, while I worked
> mainly on proving the safety of large, well-surrounded areas.
> 
	Are Thomas Wolf's papers available on the internet or in books ?

> I have worked a bit on my program since the GPW'97 paper. I can prove
> about
> 40% of the areas safe now on that test set. However, there is still lots
> of
> room for improvement. Seki is one area to work on.
> 
	I find these topics fascinating !

	Jean-Pierre Vesinet