Author: Eugene Nalimov
Date: 18:34:38 09/18/99
Go up one level in this thread
On September 18, 1999 at 19:57:18, guy haworth wrote: >Eugene, > >I'm v impressed by your 'Monday' prediction; what are the main parameters of yr >computing engine - must have loads of RAM, Hz and disc-drives. You can thank Compaq. Recently it stopped NT development on Alpha, and suddenly we have huge Alpha NT servers that have nothing to do - now they will be used only for testing service packs for already shipped products. So I was able to use 2 such servers with 4 Gb of RAM each. Unfortunately, generator itself doesn't use more than 1 CPU, and I don't want to spend time on that. But even with 1 CPU I was able to generate KQQKQQ and KRRKRR; right now they are being verified, and I think that tomorrow or on Monday I'll be able to download them. >Lewis Stiller sent me some stats which I lined up with his thesis results >successfully: I hope they might encourage and/or help you. I'll send you my statistics. It differs from Stillers, as it's DTM, not DTC. If my TBs are correct (have to wait for verifier), longest DTM in KQQKQQ is 50. Eugene >Assume White = 1st player, Black = 2nd player > >1,546,346,340 distinct KQQKQQ positions > (Ks apart but 'unreachables' possible ... > eg double-check, both sides checked) > 44 = max-depth wtm win ... and maxdepth position demonstrates > not all Q-checks lead to a win and > very easy for Black to turn a deep win into shallow win > >1,280,546,724 (= 82.81%) distinct wtm wins but lots in 1m or 2m >1,186,868,923 white wins in 1m (mate or bQ en prise) > 38,200,845 white wins in 2m (mate or wQ..bK..bK skewer) >1,271,619,037 (= 99.3% of all wins) Wh wins in 6m or less > >What depth in ply can Fritz-on-a-PC reasonably search full-width to in KQQKQQ? >I would expect it to be less than 16-ply which I have seen quoted in other >searches. The EG-db of course adds value where depths/distances beyond the >search-tree horizon are involved. > >[ If you need any more stats, plse request via this board. Got a glitch in >email at the moment. ] > > >Turning to non-KQQKQQ dbs, there is more dialogue on the GK-v-WT bulletin board >about using heuristics in a KQP(g7)KQP(d6-3) EG-db attempt ... or making >assumptions like 'the black pawns don't move'. > >The simplifying assumptions/approximations are I think: > P=Q only (so forget P=R,B,N ... if GK wins with P=N, I'll hold my hand up) > P does not take Q ... as this will reduce to an evaluated 5/4-man ending > If White has one more Q than Black for more than 1 ply, White wins. > >If GK is to win, then at the end, for some moves, Black's pawn(s) will not have >moved ... and Black could choose at some stage NOT to move them. If the Black >Ps are doing more for White than Black, White will not capture them either. > >The following suggestion, which I've seen elsewhere, therefore seems >interesting: > >"Generate the '5-man' database KQPKQ with the constraints of the Black pawns on >'bx' and/or 'dy'". This would be a new mode of EG-db generation; how this gets >implemented in retrograde analysis is not clear to me. The above three >approximations/simplifications would still apply. > >I've been offline for a while but it appears the GK-v-WT BBS has 'bottomed out' >endgame G (which has an h rather than a g pawn) and discovered that it wins for >GK. So, it looks like we are still heading for endgame-D with a g5-7 wP and >KQQKQQ as a possible outcome of that. > >See (eg) http://www.gmchess.spb.ru/kasparov-world.html where they call endgame-D >position "A". > >Rgds, Guy
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.