Passed Pawns (endgame)

[mcostalba]

You don't have to. Only $ELO makes the world go round...
Nevertheless, from a chess player's point of view this is a very stupid behauviour from SF and possibly many other engines.

I'am a very bad chess player

[lech]

All the proposed patch attempts do work well with this position. Except the patches from SF team, because, so far, they haven't shown any.

I don't have any patch to show, sorry. Actually I am not very interested in tweaking for a given position. I consider good a patch that proves to increase strength in a real games test session: I don't care how it behaves in specific positions.

Passed pawns are very important factor and need a special code. It seems to be impossible to improve it by a general code. Probably a main difference between engines is: how engine handle with the theme “passed pawns” (excluding that all are clones ). It is a pity that a possibility to improve engines is rejected. Let engines be “stupid”.

[lech]

I'am a very bad chess player

I was watching a game Fire 13 - SF 171 (4 min.+2 sec.). SF had helped Fire to get a side passed pawn (in Bishop endgame! ) and lost the game. It looked like this (I don’t remember the right side exactly).
[d] 4k3/2p2p1p/3p1bp1/1p1P4/p5P1/2P2P2/PP2K2P/2B5 b - - 0 1
SF moved 1…a3 2.b3 Bxc3 3.Bxa3 and in the next moves Fire by a2->a4 got the side passed pawn.

[mcostalba]

SF moved 1…a3 2.b3 Bxc3 3.Bxa3 and in the next moves Fire by a2->a4 got the side passed pawn.

After the above moves and ..Kd7 is not so obvious that white wins:

[d] 8/2pk1p1p/3p2p1/1p1P4/6P1/BPb2P2/P3K2P/8 w - - 0 3


I have quickly tried some engine, but result is almost draw with just a small advantage for white...

[lech]

After the above moves and ..Kd7 is not so obvious that white wins:

[d] 8/2pk1p1p/3p2p1/1p1P4/6P1/BPb2P2/P3K2P/8 w - - 0 3

I have quickly tried some engine, but result is almost draw with just a small advantage for white...

Of course, the way to win is not obvious one. But Fire, thanks to the passed pawn, methodically increased advantage and won. Sorry, I haven’t all the game. I remember only that I tested later this position and Fire proposed 1…Kd7 (P4 3.5G 2hreads). The right side (pawn structure) could be other.

[Ralph Stoesser]

Why use Max() here? tr can't become negative.
evaluate.cpp, line 892+893

Code: Select all

int r = int(relative_rank(Us, s) - RANK_2);
int tr = Max(0, r * (r - 1));

[mcostalba]

Why use Max() here? tr can't become negative.
evaluate.cpp, line 892+893

Code: Select all

int r = int(relative_rank(Us, s) - RANK_2);
int tr = Max(0, r * (r - 1));

Yes, you are right. Thanks Ralph.

[Eelco de Groot]

Why use Max() here? tr can't become negative.
evaluate.cpp, line 892+893

Code: Select all

int r = int(relative_rank(Us, s) - RANK_2);
int tr = Max(0, r * (r - 1));

I don't quite follow, relative_rank(Us, s) can be == RANK_2 if the passed pawn is still on the original rank, in which case r == 0.

What did I miss

Not that this code is still in Rainbow Serpent, I wanted to give passed pawns on their original rank also a minimal rank related bonus, so it is changed to

Code: Select all

    while (b)
    {
        Square s = pop_1st_bit(&b);

        assert(pos.piece_on(s) == piece_of_color_and_type(Us, PAWN));
        assert(pos.pawn_is_passed(Us, s));

        int r = int(relative_rank(Us, s) - RANK_2);
        int tr = Max(1, r * (r - 1));

Regards, Eelco

[BubbaTough]

I don't quite follow, relative_rank(Us, s) can be == RANK_2 if the passed pawn is still on the original rank, in which case r == 0.

What did I miss

Not quite sure I am following. Are you worrying about the result of 0 * -1 being negative?

-Sam

[Eelco de Groot]

I don't quite follow, relative_rank(Us, s) can be == RANK_2 if the passed pawn is still on the original rank, in which case r == 0.

What did I miss

Not quite sure I am following. Are you worrying about the result of 0 * -1 being negative?

-Sam

Ah, I see now, I missed that, thanks Sam, but what if a technically illegal position has a passed pawn on the bottom rank, RANK_1. In that case r really is negative and tr for RANK_1 becomes greater than for RANK_2, which is not what you really want. Not that it often would be a problem but in that case the code does not work with Max either, I agree. Maybe r has to be positive at all times that would be better, so something like tr = r * (r-1) + 1 I had been thinking about that because in my version of the code tr == 1 for both RANK_2 and RANK_3 so that is slightly illogical maybe (but bonuses are made from both tr and r * r so there still is enough difference). Not sure it would be better though. And tr for RANK_1 is still greater than RANK_2 then.

Regards, Eelco