Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Deep Blue afterthoughts

Author: Amir Ban

Date: 16:09:43 05/11/98

Go up one level in this thread


On May 11, 1998 at 09:06:05, Guido Schimmels wrote:

>>The only addition I would make to this report is a remark on the odd
>>nature of the PV for Qb6 at ply 10. GM Ronen Har-Zvi pointed out to me
>>that the move Qa7 in this PV is a simple blunder. How such a move can
>>appear in a PV is just one more mystery.
>
>The DB outputs of game 2 you've presented here, have already been
>discussed
>in an issue of the german magazine "CSS" quite some time ago.
>The author says, DB does not propagate the PV from the leaves to
>the root, as most PC-programs do. In order to save time, DB simply
>looks up the PV in the hashtable, which he thinks explains for some
>weird PV's.
>
>Guido


Yes. I haven't seen it, but this comes from Friedel, who took part in
the email discussion over this last May. The statement about how DB
finds PV originated thus: Frans Morsch saw my report and my comment on
Rxf6 and suggested this explanation. Nobody knows how how they actually
do it. I'm not arguing this point, but there is a detail in the
printouts that cast a doubt on this: In the 36th move, after the search
stopped there was an announcement "main line will be reconsructed"
presumably from the hash table (and this produced the famous axb5). But
if this is always done this way, why announce it specially on this
occasion ?

Anyway that would explain the line with Rxf6, which is obviously a false
PV and doesn't match its evaluation. The other lines are obviously true.
The ply 10 PV is a true one. It does not contain garbage but a simple
blunder.

Amir



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.