The worst pawn weakness

[Ferdy]

[d]4k3/1p1b2p1/5p1p/1p1p4/3N4/P1P5/1P3PPP/4K3 w - - 0 1

That pawn in b7 is the worst in my evaluation, it is isolated, and doubled.
It will get worser if there is white pawn in either a5 or c5
because then that b7 pawn will become backward.

That pawn in b5 is almost normal except that it is an isolated. Penalty is increased when there is no white pawn along its file in front.
It can be backward but since it is already in the 4th rank I zeroed the penalty there.

Additional scoring for piece interaction with this type of structure has a different story and is a bit challenging.
For example will white be able to create deadly outpost as a result of these weakneses? Will white be able to plant a piece
in front of the isolated pawn?

[Ferdy]

This is because white also gets the 2 bishops.

[jdart]

I am not saying don't penalize it: I am just saying the penalty, in general, should not be very high. I have done some automated tuning of this parameter so I am not just stating an opinion, but have some evidence.

My current penalty is about 1/10 pawn in the middlegame, about 1/5 in the endgame. Less if the doubled pawn is on or near the a or h file. It is harder to attack a doubled pawn on the edge.

--Jon

[Lyudmil Tsvetkov]

[d]4k3/1p1b2p1/5p1p/1p1p4/3N4/P1P5/1P3PPP/4K3 w - - 0 1

That pawn in b7 is the worst in my evaluation, it is isolated, and doubled.
It will get worser if there is white pawn in either a5 or c5
because then that b7 pawn will become backward.

That pawn in b5 is almost normal except that it is an isolated. Penalty is increased when there is no white pawn along its file in front.
It can be backward but since it is already in the 4th rank I zeroed the penalty there.

Additional scoring for piece interaction with this type of structure has a different story and is a bit challenging.
For example will white be able to create deadly outpost as a result of these weakneses? Will white be able to plant a piece
in front of the isolated pawn?

Hi Ferdinand.
Well, obviously, the interaction of terms, the sum total, is most important of all.

For example, you know that in a certain position the eval should be 50cps in favour of white. 2 different engines might choose to evaluate the 3 most important terms (comprising the essence of the position) in 2 completely different ways.

The 1st engine would have 2 bonuses of 50cps each, and one penalty of also 50cps, sum total 50cps for the eval.

The second engine would score the same terms with 2 bonuses of 25cps each, no penalty or bonus for the 3rd term. In the end you have the same 50cps for the eval.

One of the engines might have the terms completely wrong, but it will still have the right score, which is the only thing of importance. That is basically the essence of tuning: you might be able to tune the whole well while being on the wrong side with the particulars.

I read on the Stockfish development site infos that are quite consistent with this one: they have a couple of failed patches, they would then make a union of them, and, surprisingly, the union of the 3 failed patches passes the test successfully.

So, really, I think the most important thing in chess is to have a couple of spare failed patches...

[carldaman]

I suppose that doubled pawn is not necessarily a weakness per se. Penalizing it works because more often than not a doubled pawn indicates that more subtle structural weakness exists somewhere else - be it an isolated pawn, a blocked piece or pawn majority on the other wing. Probably sufficiently robust and perfectly tuned evaluation function wouldn't need a penalty for doubled pawns.

This is in line with the principle of two weaknesses. One weakness is often very difficult to convert, but it gets easier when there are two, especially on opposite flanks.

[carldaman]

I am not saying don't penalize it: I am just saying the penalty, in general, should not be very high. I have done some automated tuning of this parameter so I am not just stating an opinion, but have some evidence.

My current penalty is about 1/10 pawn in the middlegame, about 1/5 in the endgame. Less if the doubled pawn is on or near the a or h file. It is harder to attack a doubled pawn on the edge.

--Jon

It may be harder to attack double pawns on the a or h file, but the other side can often ignore such double pawns, in particular if they can be easily (or are already) blocked, to great advantage elsewhere. In my own human evaluation I regard such pawns as being no more than 1/2 a pawn, unless they greatly influence play.

CL

[Lyudmil Tsvetkov]

I think it might be useful, even with quite normal double pawns (no symmetries, no isolation) to make a distinction between a couple of categories:

[d]8/pp3pk1/1p4pp/8/5P2/2P3PP/PP4K1/8 w - - 0 1
Simply a normal double pawn, standard penalty

[d]8/pp3pk1/1p4pp/1P6/5P2/2P3PP/P5K1/8 w - - 0 1
A double pawn that is blocked by an enemy pawn; pretty much unwelcome, accentuating the weakness, making it impossible for the double pawn to undouble; some additional penalty for this might be due (1/2 of the penalty for the standard double pawn?)

[d]8/pp3pk1/1p4pp/8/2P2P2/6PP/PP4K1/8 w - - 0 1
A double pawn whose square in front is controlled by an enemy pawn; still unwelcome, might make difficult for the pawn to undouble; some additional penalty, maybe 1/4 of the standard penalty

[d]8/1p3pk1/1p4pp/p7/5P2/2P3PP/PP4K1/8 w - - 0 1
A double pawn with the more advanced of the doubled pawns being the least advanced of any friendly pawn on adjacent files; very much unwelcome, because structurally enormously deficient; some additional penalty, very big under normal circumstances, maybe 1/2 or more of the combined penalties for both double pawns

In many situations other factors might make such details go unnoticed, but in a range of other situations they might be of real importance.