Author: Chris Whittington
Date: 08:55:51 11/24/97
Go up one level in this thread
On November 24, 1997 at 09:55:34, Robert Hyatt wrote: >On November 24, 1997 at 00:47:37, Jouni Uski wrote: > >>Solution times for Mchess 6 in K6/200 harware: >> >>1. 0 sec >>2. 0 sec >>3. 17 sec >>4. 4 sec >>5. 41 sec (found Na5 after 10 ply only!) >> >>Jouni > >The first cut thru crafty found #5 quickly. It turned out to be a bug. >Here's why finding this quickly is bad: > >after giving up the knight, the opponent has two tasks: 1. Keep the >king away from the queening square because of the wrong bishop; 2. >capture >both pawns that are blocking the promotion. > >Winning the pawns can only be done at the expense of letting the other >king >reach the queening square and draw. And there is no way to capture the >pawns >and prevent this. But the program can try for a long time. > >My bug was "assuming" the king could reach a1 and draw. Which was >correct. >But there were also positions where it was just barely prevented from >reaching, >and it still thought draw. > >I'd be concerned for any program that finds this quickly, as that >program is >likely making an assumption that works in this position, but fails in >others, >where the pawn and bishop just barely prevent the king from getting to >the >queening square. This happened in crafty a couple of months back and >someone >sent me the position where it failed. It turned out that I *had* to >make sure >the king physically reached the queening square before calling it a >draw. > >So solving #5 at shallow depths is most likely wrong, because it takes >*many* >plies to actually see that black can't win the pawns and hold the king >out. So >black uses his king and bishop to drive the white king away, then runs >back over >to grab one pawn right at the horizon and thinks he is still winning. >Each >additional ply is used to extend the win of one or both pawns to right >at the >horizon without giving white a chance to reach a1... I agree. Fast solvers are either broken, or another possibility ....... They have some form of extension method to deal with the no-change-score that results from dancing with the bishop whilst holding off the king ? If someone would like to post EVALS for this postion, especially for the solvers. Negative evals mean they just choose Na5 by chance as the least bad possibility. They don't have the knowledge, but they got lucky in this position. Early draw evals means their code is broken. Draw evals with a long and valid main line mean they really understand. Chris Whittington
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.