Ending

[Ignacio]

Yes, FinalGen says 93.Bb5? is bad, but 86.Bh3 is not wrong. The position is draw.

[Lyudmil Tsvetkov]

[d]8/8/3p1k2/2pP2n1/2P2K2/3B4/8/8 w - - 0 0

It is interesting to know how could black improve on 1.Bf5 Nf7 2. Bd3 Ne5 3. Bf1 Kg6 4. Be2 with a draw.

Otherwise, blocked pawns would favour the knight, pawns on squares the colour of the bishop would also play into the hands (hooves) of the knight, but too few pawns there and too compact in structure.

Lyudmil

[Kyodai]

Yes - and typical position where a GM (IM) would play as long as possible -
just waiting (hoping) for a mistake from white. And á mistake happened
in this game...

[Lyudmil Tsvetkov]

Hi, Sune.
Thanks for posting interesting positions.
That is why chess (much more human chess at that) is not only about what really there is in a position, but about competition too, competition of nerves, personalities, idyosyncracies, etc., something that not always favours the objectively stronger player and has a negative impact on the quality of the game.

Lyudmil

[Don]

Yes - and typical position where a GM (IM) would play as long as possible -
just waiting (hoping) for a mistake from white. And á mistake happened
in this game...

Yes, from my analysis this is a draw but it's very easy for white to make a mistake. That's why you are not going to be able to "solve" this position with a deep search because there is nothing to see.

[zullil]

Yes - and typical position where a GM (IM) would play as long as possible -
just waiting (hoping) for a mistake from white. And á mistake happened
in this game...

Yes, from my analysis this is a draw but it's very easy for white to make a mistake. That's why you are not going to be able to "solve" this position with a deep search because there is nothing to see.

Ideally, positions like these should be evaluated using pre-computed endgame databases. Or so it seems to me.

[Don]

Yes - and typical position where a GM (IM) would play as long as possible -
just waiting (hoping) for a mistake from white. And á mistake happened
in this game...

Yes, from my analysis this is a draw but it's very easy for white to make a mistake. That's why you are not going to be able to "solve" this position with a deep search because there is nothing to see.

Ideally, positions like these should be evaluated using pre-computed endgame databases. Or so it seems to me.

Since the pawns are on just 2 files and will always remain so, a specialized database could cover this ending. It would be an 8 man database but with fairly heavy constraints. Probably having databases in general would help here, but the issue is to prove this is a draw and unless you have a complete database solution that might be hard to prove. I think it almost certainly is a draw but I cannot prove it.

[Lyudmil Tsvetkov]

Hi Don.
I do not understand how search functions, but I would not accept for granted the assertion that evaluation is absolutely unimportant.
When you should search more than 30 plies to see the logical development of a game, you should be aided by some practical knowledge.
One of my favourite examples concerns the advantage of the pair of bishops in the endgame.
Let us suppose that we have a position with 2 bishops and one isolated and one double pawn (part of a group of pawns) for one of the sides.
You could do 3 things:
- assign some penalty points for the isolated and double pawns and not assign a bonus for the pair of bishops, which should be absolutely wrong
- assign penalty points for the weak pawns and assign bonus for the pair, a somewhat better solution
- assign bonus for the pair and not assign penalties for the weak pawns, the right solution, as the 2 bishops would be able to help in getting rid of the isolated pawn, undoubling the double pawn, while remaining intact

In all cases the game could be decided some 20 moves later, and, in case you have chosen to evaluate in the wrong way, it would be impossible to correct things along the way.

In the same streak of thought, supposing that you already know the general rule that with 2 bishops in the endgame the pawn structure for the bishop side is irrelevant, you are faced with another position with a pair of bishops and double horizontally isolated pawns (meaning double pawns with no friendly pawns on adjacent files) for one of the sides. Going by the general rule, you assign yourself an advantage of some 50 centipawns for the bishop pair, and do not assign any penalty for the double isolated pawns. 7 moves on, suddenly your half a pawn's advantage seems to disappear somewhere, and another 7 moves on, the 2 bishops side sees itself with half a pawn disadvantage. So the tables have been turned from almost winning to almost losing in a matter of 15 moves. You might wonder what is the cause for this, but the truth is that the evaluation has been wrong from the very beginning, bearing in mind that with a pair of bishops the pawn structure for the bishop side is largely irrelevant, but maybe the only significant exception to this would be the case of double horizontally isolated pawns, that should get some penalty, worth almost a full pawn.

I do not think that all the might of the search could help, if you do not specify some rules. So that rules are important, and exceptions to the rules too:)

Concerning the second part of the above example, I am surprised that even some extremely strong engines seem to ignore it.

Best regards,
Lyudmil

[Don]

Hi Don.
I do not understand how search functions, but I would not accept for granted the assertion that evaluation is absolutely unimportant.
When you should search more than 30 plies to see the logical development of a game, you should be aided by some practical knowledge.

I think I believe in that more that most people here do. See my post where I said something similar recently and someone responded that this is a pure search problem. To me it's self-evident that most interesting positions can be played correctly without necessarily "solving" the position tactically. In practice it is not that simple of course as some positions would require fantastic knowledge to overcome the tactics which prevent it from playing correctly.

Someone a few months ago posted a locked pawn position where the only way to win was to sacrifice a queen permanently, but the winning side already had much more than a queen advantage. This is a good example of where search could eventually solve the position but knowledge could make the program play it correctly without needing a deep search.

One of my favourite examples concerns the advantage of the pair of bishops in the endgame.
Let us suppose that we have a position with 2 bishops and one isolated and one double pawn (part of a group of pawns) for one of the sides.
You could do 3 things:
- assign some penalty points for the isolated and double pawns and not assign a bonus for the pair of bishops, which should be absolutely wrong
- assign penalty points for the weak pawns and assign bonus for the pair, a somewhat better solution
- assign bonus for the pair and not assign penalties for the weak pawns, the right solution, as the 2 bishops would be able to help in getting rid of the isolated pawn, undoubling the double pawn, while remaining intact

In all cases the game could be decided some 20 moves later, and, in case you have chosen to evaluate in the wrong way, it would be impossible to correct things along the way.

With the heavy reductions current programs have the problem is exacerbated and evaluation becomes more important than ever. This is because the evaluation is in some sense guiding the search and shaping the tree.

When Komodo fails to find the right move in a position we generally blame the evaluation function. You have to distinguish between playing the right move because you can tactically resolve all the issues or playing the right move because your evaluation function correctly guides you.

In the old days computers could not see simple checkmates such a rook and king vs king, but they DID understand the basic principles to carry out this mate. A trivial example of course, but it makes the point.

In the same streak of thought, supposing that you already know the general rule that with 2 bishops in the endgame the pawn structure for the bishop side is irrelevant, you are faced with another position with a pair of bishops and double horizontally isolated pawns (meaning double pawns with no friendly pawns on adjacent files) for one of the sides. Going by the general rule, you assign yourself an advantage of some 50 centipawns for the bishop pair, and do not assign any penalty for the double isolated pawns. 7 moves on, suddenly your half a pawn's advantage seems to disappear somewhere, and another 7 moves on, the 2 bishops side sees itself with half a pawn disadvantage. So the tables have been turned from almost winning to almost losing in a matter of 15 moves. You might wonder what is the cause for this, but the truth is that the evaluation has been wrong from the very beginning, bearing in mind that with a pair of bishops the pawn structure for the bishop side is largely irrelevant, but maybe the only significant exception to this would be the case of double horizontally isolated pawns, that should get some penalty, worth almost a full pawn.

I do not think that all the might of the search could help, if you do not specify some rules. So that rules are important, and exceptions to the rules too:)

Concerning the second part of the above example, I am surprised that even some extremely strong engines seem to ignore it.

Best regards,
Lyudmil

Of course the devil is in the details. For every rule there is some exception and there is usually a cost too.