Unsolvable position

[Sidiki]

Hi Programmers

unsolvable.PNG (192.86 KiB) Viewed 5832 times

Why this position is't so difficult to solve by the bests knowed engines without endgames database?

Even with a long time of search.

Friendly,

[Joost Buijs]

Is this position won for one of both colors?
Which color has to move?

[BertTuyt]

Sidiki, in all honesty, i do not remember this position either.
When did i post it?
Could be that my age is blocking my memory

Bert

[Sidiki]

Sidiki, in all honesty, i do not remember this position either.
When did i post it?
Could be that my age is blocking my memory

Bert

, perhaps a mistake from me.

[Sidiki]

Is this position won for one of both colors?
Which color has to move?

White to move, and win Joost

[BertTuyt]

Sidiki, and what is according to you the winning move sequence?

Bert

[Sidiki]

Hi Bert,
37-31 and then 35-30


Friendly, Sidiki.

[Joost Buijs]

Hi Sidiki,

I haven't analyzed it, but it could be that the solution leads to a won EGDB position with equal material, and that finding the solution without EGDB needs a very deep search, too deep for an engine (even on very big hardware). Futility pruning and LMR (late move reduction) can also have an adversary effect on finding the winning move.

Another possibility is that the EGDB helps in finding a move sequence that leads to the win of material because other alternatives lead to a lost EGDB position for the opponent. After a 17 ply search (with extensions enabled and pruning disabled) Ares sees a win of approximately 1 disc, it doesn't find an absolute win.

Although current engines are very strong there still seems to be some room for improvement.

Joost

[BertTuyt]

Sidiki, this is indeed an interesting position.
After 37x31 27x36 and 35-30, Damage is able to find a win of only 1 piece (or as it is black to move, 1 piece lost).
I don't see an absolute win yet.

Further analysis, after the above moves, white threatens 29-24, and black can react in the following ways:

* 20-24, seems to win another piece for black, but white has a counter attack (starting with 21-26), Damage score > 2
* 19-23, also here Damage sees a score > 2, move sequence 47-42 36x47 29-24 20x29 42-37 47x33 39x28
* 9-14, here Damage sees a Database win for white, indeed a beautiful combination.
* 18-23, black sacrifices 2 pieces, so now black is overall 1 piece down and will probably lose the game.

After the initial 3 moves, Damage chooses 18-23, as this is better as a 2 piece disadvantage or a database lose. It could be that in this position another combination is possible, but Damage did not find it.

The reason Damage (and maybe other engines) have problems with this position is (as already shared with Joost) is the combination of pruning and LMR.
Due to these 2 the engines can search quite deep, as most of the sacrifices are not working. But the disadvantage sometimes you miss a combination.
This requires fine tuning of search parameters, and maybe some additional code to detect combinations and traps.

In the past Gerard Tailly was working on these types of positions if i remember well.

So we still have some work to do

Bert

[Sidiki]

Thanks for your answers,
There's a program that curiously find the absolute win even with endgame database.
I will send a video.
Friendly, Sidiki

[Sidiki]

Hi all,

This's the video link of the solved problem by Idames a private engine.
We will see the settings, no opening book, no endgame database. We hope that we will learn from each others
https://pixeldrain.com/u/TTxWDjTS

Friendly, Sidiki

[Sidiki]

This's the Android version.
Certainly as Bert mentionned it, seeing trap are implemented.

Friendly, Sidiki

[Sidiki]

Another thing,
Dam 2.2, yes you read it, Dam 2.2 of Harm Jetten find the winning sequence in less of 30 secondes when you give him 30 minutes for 75 moves.

You can do the test yourself to see.
.
Idames on computer see the solution in less of 1 seconde.

I think that as Bert and Joost wrote, it's the LMR problem

Friendly, Sidiki.

[Joost Buijs]

Solving problems needs a totally different approach, you'll see the same with chess, perfect mate solvers unable to play a strong game, and vice versa.

When I disable futility pruning in Ares, it finds the solution in a fraction of a second at 17 ply, but without pruning it starts to play weaker in normal games and that is something you don't want. It is possible to prune more selective, e.g. prune this and not that, usually this makes the program play weaker too.

Joost

[Joost Buijs]

Thanks for your answers,
There's a program that curiously find the absolute win even with endgame database.
I will send a video.
Friendly, Sidiki

Hi Sidiki,

I've been looking at the video, the program uses 5P EGDB and only finds the win of 1 disc and not the absolute win. I thought you meant that it finds the absolute win without EGDB, probably I misunderstood.

Joost