Pawn specifications

[Lyudmil Tsvetkov]

I know I should have left this forum long ago, but something continuously draws me back.

Here maybe one last thread before I leave for certain.

[Lyudmil Tsvetkov]

Well, undefended pawns certainly deserve some penalty, as they are easier targets, but, although the penalty might be unifrom in terms of ranks, it makes very much sense to distinguish between opposed and unopposed undefended pawns.

Of course, unopposed undefended pawns, undefended pawns on a semi-open file are due a somewhat larger penalty. Reason being - they are easier targets there.

[d]6k1/6p1/2p2p1p/1pP5/8/3P1PP1/P1P4P/6K1 w - - 0 1

above, h2,f3,c2,c5 and a2 are undefended pawns for white, of those only a2 is unopposed, on a semi-open file, and thus due a bigger penalty.

for black, g7 and c6 are undefended, but both are opposed, so normal penalty in this case

People think undefended pawns are unimportant and they might not be flagged in terms of opposed/not opposed, but that is not precisely so. Specifying always helps.

[Lyudmil Tsvetkov]

[d]6k1/2p4p/2p1p3/8/P2PP3/2P3P1/7P/6K1 w - - 0 1

OK, above a4 is a passer and an unopposed pawn, g3 is a candidate passer and an unopposed pawn, and both are scored in most good engines, so scoring unopposed pawns in the general case will only lead to redundancies.

But what about the d4 pawn?
It is neither a passer, nor a candidate passer, and yet it is unopposed.
Does not this pawn deserve some additional bonus?

Of course, it does, as all unopposed pawns are bonused for their ability to advance easier than if they had been opposed by enemy pawns. The bonus might be small, some 2-3-5-8cps, maybe uniform, but it is still due.

Well, the most important unopposed pawns, like passers and candidate passers already have their nice bonus points, but why leave a true existing chess term unattended?

How can we resolve the problem?

Well, very simple: assign the small bonus to any unopposed pawn, while excluding passers, candidate passers and backward pawns from the bonus.
Of course, backward unopposed pawns are even worse than backward opposed, so giving bonus here is also strictly unproductive.

[d]6k1/2p3p1/1p5p/1P5P/8/8/6P1/6K1 w - - 0 1

of course, no one would like to give the bonus to the c7 backward pawn above, as its condition is even worse than that of the g7 backward pawn

So, I think, if you would like to have such a term in the general case, that could be possible, if the bonus is given to any unopposed pawn that is not either a passer, candidate passer or backward pawn.

And, rest assured, there are many unopposed pawns falling into neither of the above 3 categories that are eagerly waiting for their small due bonus.

[Lyudmil Tsvetkov]

Well, this is about pieces, but, as pawns are also involved in the rule, it could easily pass as a pawn specification.

It is very reasonable to give larger bonus, maybe by 1/10 or so, to any outposted minor that is blocking an enemy undefended pawn. Of course, such outposts are multifunctional - the pieces are outposted and at the same time stop the possible advance of an enemy weak pawn, making it even weaker and an easier target.

Apart from that, the enemy pawn would shelter the outposted piece from frontal attacks by enemy heavy pieces.

[d]6k1/8/5p2/2p2Bp1/2N3P1/8/8/6K1 w - - 0 1

above, Nc4 and Bf5 are such minor outposts, blocking an enemy undefended pawn, deserving a bigger bonus.

Nc4 blocks an enemy isolated pawn, while Bf5 blocks an enemy backward pawn, but, as both the isolated and backward pawn are undefended, you might simply specify outposts blocking undefended pawns.

As mentioned, such outposts enjoy additional advantages and therefore the bonus.
In any case, they are much preferable to minor outposts that do not block enemy undefended pawns.

As, in a well-tuned engine, the rule might not quite work with simply adding it, it might make sense to slightly decrease the general outpost bonus, by maybe 1/20, and only after that give the additional 1/10 increase specifically to outposts blocking enemy undefended pawns.

Blocking an enemy pawn would mean that the pawn is immediately in front of the minor.

[nionita]

Well, the most important unopposed pawns, like passers and candidate passers already have their nice bonus points, but why leave a true existing chess term unattended?

But I have a question here: opposed pawns anihilate each other. Then if we have for one side more unopposed pawns, that side must have more pawns at all. Is this in an end game not already a bonus, as the pawn value increases?

[Lyudmil Tsvetkov]

Just an idea.

I am watching SF play some positions that are supposed to be won, but, when it should gain decisive advantage, in many cases it simply fails to decisively increase its advantage by specifically pushing storming pawns that also belong to the own king shelter.

The win is there, SF only needs to push its storming pawns belonging to the own king shelter a bit further, and the advantage will increase, yet, it does not do so.

I wonder why?

And when I think about that, a stupid thought crosses my mind - when picking moves for search, why not specifically prefer moves with the pawn or piece's destination square being one of the rightmost e,f,g,h files?

I do not know, maybe someone does this, but, in any case, that makes very much sense, as about 2/3 of all meaningful moves, at least in the mg, eg could be excluded from that, involve pawns or pieces whose destination square is on the right part of the board - one of the e,f,g,h files - rather than the left part, the a,b,c,d files.

When you have such a predominance of meaningful moves landing to the right, why keep the symmetry in move-picking and search?

Of course, the asymmetry here would be due to the fact that the most important moves are associated with attacks on the enemy king, stroming pawns and transferring pieces closer to the enemy king. As it seems that only bonusing eval is not quite sufficient to resolve all issues, why not simultaneously also bonus move-picking and search, of course only for the mg, where attacks are prevalent?

As Kasparov rightly noted somewhere, why bother with other issues, when there are more important concerns, like king safety?

Anyone already having right-left move-picking asymmetry in their engine, or maybe everyone is already doing this, and I simply have not noticed it?

[bob]

I do this a bit differently in Crafty. "unopposed" for some pawns is bad, like the backward c pawn you gave, but it is not bad because it is unopposed, unless the opponent has a rook or queen to attack it from the front. Otherwise a backward pawn is simply backward. I count these weak pawns that are unopposed, but I give the penalty in the piece scoring where a queen or rook gets a bonus if the other side has such pawns...

[mjlef]

I know I should have left this forum long ago, but something continuously draws me back.

Here maybe one last thread before I leave for certain.

Lyudmil,

Larry and I both enjoy your postings. Your ideas are very stimulating, so please keep posting. I cannot say you always agree with all your ideas, but most of them end up starting discussions and often lead to things that make our program stronger.

Mark

[Dann Corbit]

I know I should have left this forum long ago, but something continuously draws me back.

Here maybe one last thread before I leave for certain.

Your posts are so interesting that if you never existed, then someone should have invented you.

The problem is that we don't have the technical expertize to do it. So if you go, it will make a hole.

[Steve Maughan]

I know I should have left this forum long ago, but something continuously draws me back.

Here maybe one last thread before I leave for certain.

Lyudmil,

Larry and I both enjoy your postings. Your ideas are very stimulating, so please keep posting. I cannot say you always agree with all your ideas, but most of them end up starting discussions and often lead to things that make our program stronger.

Mark

+1 Totally agree!

As Mark said, it's good to have someone like you who takes a topic and distills it down to the fundamentals.

Steve