A possible new pawn structure evaluation term?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Arpad Rusz
Posts: 270
Joined: Sat Apr 17, 2010 12:34 pm
Location: Budapest

A possible new pawn structure evaluation term?

Post by Arpad Rusz » Thu Oct 15, 2015 9:53 am

Tonight I was watching the Stockfish-Komodo game from TCEC when this position was reached:

[D]1r2r1k1/1b3pp1/1q1p1nn1/3P2N1/1bp1PPPp/2N4P/1BQ3B1/4RR1K w - - 1 27

Stockfish showed a confident evaluation of 0.79, probably thinking that he will be a pawn up after he manages to capture the c-pawn. Few moves later it took that pawn.

[D]r3r1k1/5pp1/q2p1nn1/3P2N1/1bb1PPPp/2N2R1P/1B4B1/2RQ3K w - - 11 36

An interesting feature of this position is that all pieces are still on the board but almost all pawns from the queenside have already dissappeared leaving a pawn structure with only a small width (files d-h). The evaluation dropped drastically: Stockfish probably have realized that the possible endgames after massive exchanges were all drawish... Now I was safe to go to sleep without fearing that I will miss something. :wink:

Maybe already in the middlegame the size "minimum bounding rectangle" (here d3-h7) of the pawn entire structure can be some measure of the drawishness of the possible endgames? Of course the height of that rectangle is a less important factor than the width but probably it has also some importance (the more close are the opposing pawns to each other the more likely are some further pawn exchanges.)

What do you think? Does it makes sense?

User avatar
cdani
Posts: 2104
Joined: Sat Jan 18, 2014 9:24 am
Location: Andorra
Contact:

Re: A possible new pawn structure evaluation term?

Post by cdani » Thu Oct 15, 2015 10:04 pm

At first sight I will say that seems more a problem about the pawns that are weak, and can be forced to advance and leave holes, or in the long term cannot be defended.

bob
Posts: 20406
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: A possible new pawn structure evaluation term?

Post by bob » Fri Oct 16, 2015 4:26 am

Arpad Rusz wrote:Tonight I was watching the Stockfish-Komodo game from TCEC when this position was reached:

[D]1r2r1k1/1b3pp1/1q1p1nn1/3P2N1/1bp1PPPp/2N4P/1BQ3B1/4RR1K w - - 1 27

Stockfish showed a confident evaluation of 0.79, probably thinking that he will be a pawn up after he manages to capture the c-pawn. Few moves later it took that pawn.

[D]r3r1k1/5pp1/q2p1nn1/3P2N1/1bb1PPPp/2N2R1P/1B4B1/2RQ3K w - - 11 36

An interesting feature of this position is that all pieces are still on the board but almost all pawns from the queenside have already dissappeared leaving a pawn structure with only a small width (files d-h). The evaluation dropped drastically: Stockfish probably have realized that the possible endgames after massive exchanges were all drawish... Now I was safe to go to sleep without fearing that I will miss something. :wink:

Maybe already in the middlegame the size "minimum bounding rectangle" (here d3-h7) of the pawn entire structure can be some measure of the drawishness of the possible endgames? Of course the height of that rectangle is a less important factor than the width but probably it has also some importance (the more close are the opposing pawns to each other the more likely are some further pawn exchanges.)

What do you think? Does it makes sense?
I would think many implement this in one way or another, as pawns on both wings is good if your position is better, but worse if you are struggling to draw.

User avatar
cdani
Posts: 2104
Joined: Sat Jan 18, 2014 9:24 am
Location: Andorra
Contact:

Re: A possible new pawn structure evaluation term?

Post by cdani » Fri Oct 16, 2015 6:53 am

Going to an endgame, we saw already in TCEC two games of Gull knowing that KRPPPKRPP with pawns on the same side was near draw, and Komodo and Stockfish entering it thinking to win because of near +1.

It's because of this function on Gull:
template <bool me> int krpppkrppx() {

bob
Posts: 20406
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: A possible new pawn structure evaluation term?

Post by bob » Fri Oct 16, 2015 6:23 pm

cdani wrote:Going to an endgame, we saw already in TCEC two games of Gull knowing that KRPPPKRPP with pawns on the same side was near draw, and Komodo and Stockfish entering it thinking to win because of near +1.

It's because of this function on Gull:
template <bool me> int krpppkrppx() {
I did say "many" and not "all". :)

However, that being said, I think the basic idea of the "span" of pawns is pretty important. IE if you are a pawn down, but all pawns are on 1/2 of the board, that's tough to win. As the span increases, winning chances go up since pieces and the king have to defend both sides of the board rather than just one.

Bob

Post Reply