Pawn value estimation

[Dann Corbit]

Is there any scientific measurement that yields reliable pawn value estimation?

It seems that the standard methods are very good at showing the relative values of pieces (compared to one another), but not pawns.

[jordanbray]

IMHO, it's not quite that simple. Base pawn value can change based on whether it's a rook pawn or not, for example.

More importantly, there are many other positional ideas that relate around pawns, such as structure, passed pawns, open files, isolated pawns, doubled pawns, so such values would need to be computed on a per engine basis, and what works for stockfish would not work for me.

Also, most of those methods to compute the values of the pieces hold the value of a pawn at a constant, because if they didn't then there would be many possible solutions (for example, multiply everything by 10).

[hgm]

Indeed. The practical value of a Pawn depends so much on what kind of Pawn it is (e.g. 2nd-rank isolated Rook Pawn vs connected 7th-rnk passer) that the question is ill defined.

[Dann Corbit]

Indeed. The practical value of a Pawn depends so much on what kind of Pawn it is (e.g. 2nd-rank isolated Rook Pawn vs connected 7th-rnk passer) that the question is ill defined.

Yes, but those are positional parts.

In the same way, a knight outpost close to the opponent king can be worth a rook. But those weights can be computed separate from the "wood count".

Two opposite colored bishops are stronger than their material count supposes, especially when the board gets sparse.

Put a rook on the 7th, he gains weight. Pair him with a buddy and he suddenly flexes bigger muscles.

And yet we can get stable weights for the rooks and knights and bishops.

[Dann Corbit]

IMHO, it's not quite that simple. Base pawn value can change based on whether it's a rook pawn or not, for example.

More importantly, there are many other positional ideas that relate around pawns, such as structure, passed pawns, open files, isolated pawns, doubled pawns, so such values would need to be computed on a per engine basis, and what works for stockfish would not work for me.

Also, most of those methods to compute the values of the pieces hold the value of a pawn at a constant, because if they didn't then there would be many possible solutions (for example, multiply everything by 10).

The absolute numbers are irrelevant to me. I am interested in the ratios.
All chessmen have positional terms, even the king.
All the ratios seem stable except for the ratios of pawns to other chessmen.

Perhaps the reason is that the positional terms are a larger percentage of their value than the positional terms for other chessmen.

[lucasart]

Is there any scientific measurement that yields reliable pawn value estimation?

It seems that the standard methods are very good at showing the relative values of pieces (compared to one another), but not pawns.

By definition, the value of a pawn is 100cp.

I don't understand the point of your question. Simply pawn=100 (or rather pawn value endgame = 100) is a fixed point, and all other piece values (and any eval terms) are relative to that value.

[michiguel]

Is there any scientific measurement that yields reliable pawn value estimation?

It seems that the standard methods are very good at showing the relative values of pieces (compared to one another), but not pawns.

Yes, a pawn unit is the positional value in which the side with that advantage will have a statistical performance of 64% against an opponent of equal strength.

Miguel

[Dann Corbit]

Is there any scientific measurement that yields reliable pawn value estimation?

It seems that the standard methods are very good at showing the relative values of pieces (compared to one another), but not pawns.

By definition, the value of a pawn is 100cp.

I don't understand the point of your question. Simply pawn=100 (or rather pawn value endgame = 100) is a fixed point, and all other piece values (and any eval terms) are relative to that value.

Engines tend to use other values (see, for instance Stockfish)
PawnValueMg = 198, PawnValueEg = 258,

Though pawns are generally reported to the user as 100 centipawns.

Be that as it may, I am talking about the ratio of the value of a pawn to the value of other chessmen.

The fitting methods arrive at very similar ratios for the value of any two pieces, using many different input data sets, if there are enough games.
However, the ratio of a pawn to the other chessmen varies wildly.

[lauriet]

Does this matter ?
Computers now beat the best human player, so the current programs must have it pretty right.

[hgm]

Yes, but those are positional parts.

In the same way, a knight outpost close to the opponent king can be worth a rook. But those weights can be computed separate from the "wood count".

Two opposite colored bishops are stronger than their material count supposes, especially when the board gets sparse.

Put a rook on the 7th, he gains weight. Pair him with a buddy and he suddenly flexes bigger muscles.

And yet we can get stable weights for the rooks and knights and bishops.

The problem is that the 'positional parts' for Pawns can be much larger than those of Knights, Bishops or Rooks, while their bease value is at least 3 times lower. So the 'positional part' of the value becomes larger than the base value itself. This means you can no longer extract the latter from empirical data without accurately correcting for the other first. Otherwise it would be completely eclipsed.

In a sense almost all the value of a Pawn is positional. Tactically a Pawn might be worth as little as 25cP. The remainig part is just the promise of promotion, and the ability to prevent promotion of opponent Pawns. Just measure how much an extra non-promotable Pawn is worth (e.g. delete f7, and replace f2 by a non-promotable Pawn), and you will see it is almost nothing.

Pawns are like the photons of Chess: they have no rest mass!