Natch switch (was Re: [Retros] Two little SPG-challenges)

Mario Richter mri_two at t-online.de
Wed Nov 17 10:37:23 EST 2004


Hello,


> Mario wrote:

> M> I would wish the programmers of Natch & Co. would provide a

> M> switch, which forces the detection of shorter solutions. Is

> M> there any specific reason not to do so?



> Noam wrote:

> N> I don't have any first-hand experience with Natch, but I

> N> imagine that for most SPG's in N, once Natch has solved it in N

> N> moves it would still take only a fraction of the time to check

> N> for solutions in N-0.5 or N-1.0, and a tiny fraction for even

> N> shorter solutions. For all I know, it might not even save that

> N> much time for this search to be done together with the

> N> full-length solution.


whereupon Pascal replied:

P> I don't think that such a switch would be usefull. For most SPG's in
P> N moves, if they have solutions in fewer moves, there would be many
P> cooks in N moves.

I agree with the latter (even though the example composed by Gerd Wilts
(PDB P0002233) shows that this statement is not always true) -
but finding those cooks can take a large amount of time too.

P> I think that finding all solutions, even shorter ones, in one time
P> would take much less time than checking it in several steps. But this
P> difference would probably not be of such importance to a human, as
P> Noam points out.


>From your answers it seems that both Noam and you think that I

want to save some time by getting shorter solutions on the fly.

But my idea was quite the opposite:
Saving time by stopping a probably much time comsuming search
for an SPG in N as soon as a shorter solution in N-k is detected.
(i.e the semantics of the switch would be something like
"stop if shorter solutions are detected" if set, otherwise the
usual behavior).

P> An exception may be for problems from the "Slaughterhouse",
P> where Natch is quite weak. (But in this case it is
P> probably better to use a program complementary to Natch : Popeye.)

I used popeye (which was said to be useful for those massacre SPGs)
to investigate a K+K position with 36 plies.
After 3 weeks of continuous computation on a rather powerful machine
a first solution was provided, after some more time more solutions
followed.

The idea, that such a switch might be useful, was born after F.Labelle
published his results for the K+K-35 plies case and I found my
position in his results set :-).

P> So, I don't think I will include such an option in Natch. But, as I
P> always tell to people asking for improvments in Natch, sources are
P> publicly available, and anybody is welcome to modify them to suit
P> their needs.

To summarize:
There is no *technical* reason for not introducing such
a switch, it's only that you consider it not useful?


Thanks, especially to Pascal for providing 'natch'!

mario

P.S.

The aforementioned problem by Gerd Wilts

> from Die Schwalbe 1994, recently awarded a 3rd Honorable Mention

> rn1q1b1r/ppp1pbpp/4pp2/1k5n/7P/4P3/PPPPKPP1/RNBNQB2

> a) SPG in 10.0

> b) PG in 12.5


was - according to the comment in the award - the first counter-example
to the thesis, that the difference k for a position, which has both a
unique PG in N and in N+k moves, can't be larger than 2.

Does somebody know what the current record for this difference is?






More information about the Retros mailing list