1.g4 opening is losing?

[Alayan]

The start position with queen odds is a forced loss for the side without the queen.

It's also impossible to provide a solution game tree that demonstrate a mate in X.

[jp]

queen odds is a forced loss

Yes, so we have a formal belief, but not an informal proof.

[Ovyron]

The start position with queen odds is a forced loss for the side without the queen.

I think Zullil would call that a "conjecture"!

I always wondered what would be the smallest material odds that could be had for a forced loss, I had never expected equal material to allow it with just a lousy move like 1.g4

Is chess still draw with perfect play with a2 pawn missing? Is 1.g4 so bad that you'd rather start without a pawn than with 1.g4??

[Alayan]

A very bold conjecture, even !

Coming up next : correspondence games with the a2 pawn missing to see if white can draw.

I think white has better chances without the a2 pawn than after 1. g4.

[jp]

The start position with queen odds is a forced loss for the side without the queen.

I think Zullil would call that a "conjecture"!

It is a conjecture, but a very boring one, because it's everyone's formal belief, but no one can prove it informally, so there's nothing to say...

[jp]

Coming up next : correspondence games with the a2 pawn missing to see if white can draw.

Dann had a list of five bad openings to get through first.

[Ovyron]

It is a conjecture, but a very boring one, because it's everyone's formal belief

Not everyone's, sometimes a Knight is enough!

[Alayan]

It is a conjecture, but a very boring one, because it's everyone's formal belief, but no one can prove it informally, so there's nothing to say...

You're mixing up the meaning of formal and informal.

[jp]

You're mixing up the meaning of formal and informal.

No, I'm actually not. You have to have read the posts before (from p.14, or p.13 even) to know where this came from. (But making up the term "formal belief" was not entirely serious, though it seems okay; at least it does not appear to have a technical meaning already, which the other term does have.)

[Zenmastur]

Too many possibilities to prove it but what could be feasible is to test all white responses within let's say 0.2 pawns off the main SF and LC0 line (more lines closer to the root) down to >3.0 eval at decent time controls. This would pretty strongly suggest that it's lost. You could start at just 0.1 and 2.0 and make it a distributed project. Same with some other opening lines. You could also try to get an actual estimate of the likelihood that 1. g4 is lost with perfect play but that's for another post.

Here's something to bear in mind:

[d]8/6pk/7r/8/8/7P/6QP/7K w - - 0 1

Without 7-man endgame tables, Stockfish-dev's (static) evaluattion of this position is +4.67. But it's a draw. How many similar positions are there, say with eight or nine men, that Stockfish totally misevaluates?

So in the middlegame, when Stockfish is searching at selfdepth 80, it's making choices that are influenced by completely erroneous leaf evaluations.

What if the only way to draw with 1. g4 is to head for such a position? Will Stockfish find it?

The static evaluations of current engines in endgame positions (not covered by tablebases) is very very far from perfect. And if they can't correctly evaluate endgames, they can't evaluate certain earlier positions either.

Code: Select all

./stockfish 
Stockfish 060220 64 BMI2 by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
position fen 8/6pk/7r/8/8/7P/6QP/7K w - - 0 1
d

 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | p | k |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | r |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | P |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | Q | P |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | K |
 +---+---+---+---+---+---+---+---+

Fen: 8/6pk/7r/8/8/7P/6QP/7K w - - 0 1
Key: 03735D4FB6B3B187
Checkers: 
eval
     Term    |    White    |    Black    |    Total   
             |   MG    EG  |   MG    EG  |   MG    EG 
 ------------+-------------+-------------+------------
    Material |  ----  ---- |  ----  ---- |  6.47  6.61
   Imbalance |  ----  ---- |  ----  ---- |  0.19  0.19
       Pawns | -0.22 -0.66 | -0.08 -0.20 | -0.14 -0.46
     Knights |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
     Bishops |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
       Rooks |  0.00  0.00 |  0.10  0.02 | -0.10 -0.02
      Queens |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
    Mobility |  0.41  0.67 |  0.15  0.73 |  0.26 -0.06
 King safety | -0.42 -0.05 | -2.01 -0.40 |  1.59  0.35
     Threats |  0.10  0.10 |  0.42  0.42 | -0.32 -0.32
      Passed |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
       Space |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
  Initiative |  ----  ---- |  ----  ---- | -0.58 -0.82
 ------------+-------------+-------------+------------
       Total |  ----  ---- |  ----  ---- |  7.38  5.48

Total evaluation: 4.67 (white side)

I've had positions like this come up in my analysis of CC games. When I can't make progress I simple avoid the line of play that leads to that position. It's actually pretty easy to do because I'm generally looking 100 plies deep which gives me 50 decision points in which I can select a different line of play.

The static evaluations of current engines in endgame positions (not covered by tablebases) is very very far from perfect.

But, does that matter? Because an user wouldn't just take Stockfish's eval as gospel and play the highest scoring move, they'd check if the engine can still produce moves that lead to a won position from that score. The winning player is expected to have a line that goes up to a won position (and I mean a won position, not one with high score) and then if the other side can't avoid reaching it or the avoidance leads to other won positions, we know it's won without needing to prove it.

We don't care about what eval Stockfish gives to the moves because the user has other ways to check if the positions are won, they can be won as low as -1.8 eval so if the user knows it they'd rather reach this position than a 3.00 one where the result is unknown.

You're right, if you're doing a good job that doesn't matter. There are other ways to tell if you are making progress. If you're not making/can't make progress it's time to try other lines.

You don't need perfection to win won positions,

It depends how difficult the theoretical win (or draw) is.

Right. Is there a handy example of a 7-man position that is a theoretical draw for the side to move, but for which only one or two non-obvious moves hold the draw? Imagine how many such positions there would be with 8 men. Or 11 men.

It's very hard to prove a negative i.e. a draw. It's neither a win nor a loss. On the other hand it's entirely possible to prove a win/loss. So, I'm not sure what point your making.

I would note that this position and much more complicated ones like it can be solved up to 10-12 man with current hardware and software. I know because I've done it. “Like” means only one piece per side and the rest are pawns. It's entirely possible to use SYZYGY 7-man TB's and a multi threaded program similar in function to Finalgen to generate the solutions to much more complicated positions with current hardware.

Right. Is there a handy example of a 7-man position that is a theoretical draw for the side to move, but for which only one or two non-obvious moves hold the draw?

I'm interested in finding such positions too, e.g. the simplest possible endgame positions that are too hard for computers alone or even centaurs. They'd probably need to be at least 5-man, I guess.

[d]6N1/3n4/3k1b2/8/1r6/5K1Q/8/8 w - - 0 1

Apparently only one move wins. Good luck to all the centaurs. And to all the engines without endgame tables.

First, I think we need to separate endgames into two classes, those with pawns, and those without.

Those without have to be a win in 50 moves or less or a piece must be captured, or it's a draw by definition. This limits the scope of the search considerably. It's still a HUGE state space to search but it does make the problem more tractable.

The position above is a draw I.e a cursed win. I would never walk into such a position in a real CC game unless I knew the outcome before hand. And neither would any other CC player worth their salt. I.e these types of positions do not magically appear on the board. You have to play into them. Why would anyone do that unless the knew the outcome before hand.

I have had endgame positions more complex than this in my analysis of real games. I.e with many pawns and unbalanced pieces. I couldn't prove a win ( I was the superior side) so I just avoided the exchanges that lead to the unsolvable position. It's pretty easy to do in practice.

yes, exactely @ Louis

or even worse if you like:
[d]b4bN1/4p1p1/1Q1p2K1/q1np4/3p1Rnk/6pr/4Rppr/3N4 w - - 0 1
mentioned by Corbit afaic remember
i gave SF11 15 min with 4 cores+TB and it shows 1.Qxa5 with a score of about -46...time to resign, isnt it?
well you unmisunderstandably sense the irorny

and as sayed, there are countless of such positions oc

I'm not sure of your point. My machine solves this in less than 2 minutes.

The question is still if those examples are relevant, because the winning player can just tag the position as drawn (who cares if it wins if I can't find the win?) and go for a different won position that is clear.

The job of the winning player is not to maximize the engine's eval or play the fastest way to mate, their job is to play into positions that are the easiest to win, and "only one move wins" positions would be hard to win by definition, and could be just tagged as draws and avoided.

And how will you even recognize such positions? Say those with 11 men? Or 19?

Any position in which you can't make progress gets tagged as such. No great mystery. If you can't make progress in a position don't play the line that leads to it. It's as simple as that!

Right. Is there a handy example of a 7-man position that is a theoretical draw for the side to move, but for which only one or two non-obvious moves hold the draw?

I'm interested in finding such positions too, e.g. the simplest possible endgame positions that are too hard for computers alone or even centaurs. They'd probably need to be at least 5-man, I guess.

[d]6N1/3n4/3k1b2/8/1r6/5K1Q/8/8 w - - 0 1

Apparently only one move wins. Good luck to all the centaurs. And to all the engines without endgame tables.

Stockfish (with 6-man tables) has Qf5 with eval +0.38 at depth 54. Wrong move, and eval is off by infinity!

Clearly I need to find a "centaur" to help.

Qf5 produces a draws just like Kg2 does. Remember this is cursed win i.e. a DRAW by definition.

Without 7-man endgame tables, Stockfish-dev's (static) evaluattion of this position is +4.67. But it's a draw. How many similar positions are there, say with eight or nine men, that Stockfish totally misevaluates?

Well yeah, that's why I was saying we can get a probability, not a certainty. It wouldn't be "solved" but it'd be something like 99.9% sure it's losing. The chances of such positions occurring in real games are very, very low. BTW, LC0 without EGTBs evaluates it as around 0.2 right away.

I wonder if there is a larger or a small percentage of such positions going up to 8, 9, 10-piece EGTBs. I'm guessing smaller. We could take random 5-piece and 6-piece EGTBs positions to see where programs without EGTBs misevaluate compared to EGTBs and it would probably be the same ratio as when going up to more pieces.

No need to test random positions. There are tables that will tell you how many positions are won and in how many moves, how many of these are cursed wins, losses or cursed losses. Simply add them up and divide by the total positions in that file and you will have your answer. Easy peasy. For pawnful endgames it would be “nice” to have a break down by how advanced the pawns are. This however, would be a lot of work, but less than it took to create the tables, and it would be quit useful.

You might want to randomize in a way that would correspond to real positions occurring in games also (e.g. by playing out games with a bit of randomness or using games databases and getting the needed positions).

Yes, I'm not sure how to weight it towards "common" positions, because who's to say what's more common in human (or computer) endgames is more common in higher-level chess?

For the test, I wonder whether the depth or nodes you allow the engines should be a function of the number of pieces. Engines may not have a problem with 4-man positions, but that's partly because they can calculate everything through to the end.

It would be interesting to know just how well/badly Stockfish statically evaluates rook and pawn endgames.

For example, this one is a draw, but Stockfish's static eval is +1.46. Imagine what might happen with, say, 9-man rook and pawn positions. And these wrong evaluations might "poison" the evaluations of all sorts of middlegame positions.

[d]8/4kp1R/8/4KP1P/8/1r6/8/8 w - - 0 1

Code: Select all

./stockfish 
Stockfish 060220 64 BMI2 by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
position fen 8/4kp1R/8/4KP1P/8/1r6/8/8 w - - 0 1
d

 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | k | p |   | R |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | K | P |   | P |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   | r |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+

Fen: 8/4kp1R/8/4KP1P/8/1r6/8/8 w - - 0 1
Key: D3383250DDB43FBF
Checkers: 
eval
     Term    |    White    |    Black    |    Total   
             |   MG    EG  |   MG    EG  |   MG    EG 
 ------------+-------------+-------------+------------
    Material |  ----  ---- |  ----  ---- |  0.01  1.29
   Imbalance |  ----  ---- |  ----  ---- |  0.00  0.00
       Pawns | -0.11 -0.27 | -0.02 -0.07 | -0.08 -0.20
     Knights |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
     Bishops |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
       Rooks |  0.00  0.00 |  0.22  0.12 | -0.22 -0.12
      Queens |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
    Mobility | -0.01  0.38 |  0.27  0.80 | -0.28 -0.42
 King safety | -0.63 -0.24 | -0.38 -0.05 | -0.24 -0.19
     Threats |  0.34  0.49 |  0.32  0.28 |  0.01  0.21
      Passed |  1.11  1.32 |  0.00  0.00 |  1.11  1.32
       Space |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
  Initiative |  ----  ---- |  ----  ---- | -0.06 -0.29
 ------------+-------------+-------------+------------
       Total |  ----  ---- |  ----  ---- |  0.25  1.61

Total evaluation: 1.46 (white side)

What?

Who cares about it's static evaluation. Everyone knows, or should know that all static evaluations are flawed. If they weren't they would be know need to do any searching. PERIOD!

As it is SF finds the right move on ply one without the use of TB's. It gives a evaluation of 0.00 in less than 3 seconds on my machine at depth 39.

And how will you even recognize such positions? Say those with 11 men? Or 19?

Here's a 20men position where black wins by force:

[d]3r2k1/1Br2ppp/1p6/p5P1/2bNP3/n1P1BP2/7P/R5K1 b - -

If you don't know how do I know no software or hardware would help you

I doubt you know. I have no doubt you think you know.

We'll come back to this one when I don't have my comp tied up doing other things. Needless to say I think, I can get SF to show a mate in "x" from this position. But I will need a few days when I can work on it without having to do other things. It looks to be a mate in 60 to 70 moves (excluding the TB part. i.e. 60 to 70 moves before it converts to a purely TB position). It's clearly a won position for black. It's just a question of do I have enough memory to get SF to see it. So far the longest mate I've been able to get SF to see is a mate in 56 (all the way to mate no TB's involved). This position will be a good test. It most likely will convert to a KBNvK endgame. But with 6 or 7 man TB's it may be solvable.

There won't be any proof tree because SF doesn't produce those and I don't have any software that does. Just SF analysis that either gives a mate or some that ends in TB score.

To be continued....


Regards,

Zenmastur