Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Can Rybka Cluster or any other engine consistently defend?
- - By Alkelele (****) Date 2013-09-14 23:20
In this post there is an interesting endgame position from the Berlin defense:

r1k1r3/ppp5/4b2p/2N1P3/1n3PP1/1P5P/P2R4/5RK1 b - -

I wonder if the position may be an example of a position possessing the following 4 characteristics:

1) It is conceptually simple (white either queens or not) unlike, say, the starting position.
2) A fairly shallow engine search gives white the "default" white advantage of about 0.10-0.15.
3) It is impossible to work out by search for sure if the position is a draw or a win for white.
4) The position is drawn, but no engine of today will be able to consistently draw it with black even if given, say, 24 hours per move (because drawing requires too much precision for the engine to handle).

This set of characteristics is crucial because if such positions exist, then it will be much harder to create a, practically speaking, unbeatable chess entity.

The bet to create such an entity hinges on two things:

1) An engine, when given a fixed amount of search time of, say, 10 minutes on one core, will never give a conceptually simple and losing endgame position a 0.10 score.
2) Once the engine is in a simple 0.10 endgame position, it will be able to draw it 100% of the time when given, say, 24 hours a move.

When I write it out, number 1 does seem rather far fetched. So, is anybody aware of somewhat simple and losing endgame positions where state of the art engines incorrectly gives a drawish 0.10 score? I guess that corr players should have some feeling about this!?

Here is analysis of the position with a fairly shallow Houdini 3 search in 4-variation mode:

Then I made a 7 hour search, and reached values around 0.20. But when I went to split points in IDeA and used deep IA searches, I ended up with:

It does seem white is making ever so slow progress when I look at the PVs. Or at least, something does happen, whether it favors white or not.

Nelson, you must have at least 4 games with this position (I would venture a guess of 20)?

Nick, up for a Monte Carlo?

What say you all, draw or win for white?

Maybe the position is not that interesting, but it has possessed me now.
Parent - By Nelson Hernandez (Gold) Date 2013-09-15 00:34
11 games. 

Bf7 played once, black lost.
Bg8 played ten times, six draws, four black losses.

Book also contains b6 as a never-played novelty.
Parent - By Nick (*****) Date 2013-09-15 00:59

> Nick, up for a Monte Carlo?


500 (tc 1s+1s) games Stockfish v Stockfish

White 59.4%, draws 77.6%, Black 40.6%

500 (tc 1s+1s) games Houdini v Houdini

White 56.6%, draws 68.0%, Black 43.4%

In the vast majority of these fast (the depths reached are shown) games, black starts with Bf7 which is not your IDeA-found optimal move.  Running lots of fast games can be useful to seed IDeA with, but getting the best results from them takes a little management, I've found.  I usually run them at a point where IDeA can't decide on a single best move.
Attachment: position.7z (456k)
Up Topic Rybka Support & Discussion / Rybka Discussion / Can Rybka Cluster or any other engine consistently defend?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill