Without the null move observation

[Barardo]

Hi everyone.
I have a question: is it possible to create a 2800 elo (theoretical) chess engine without using the null move observation?
If so what would be the algorithms that it would use?

[ZirconiumX]

Possibly.

Personally, I think (Multi)ProbCut or MultiCut may be used, but these tend to hurt engine playing strength and tactics.

Late Move Reductions, probably Move Count Based Pruning is a must.

Then Futility Pruning, ectc.

Matthew:out

[bob]

Hi everyone.
I have a question: is it possible to create a 2800 elo (theoretical) chess engine without using the null move observation?
If so what would be the algorithms that it would use?

Should be doable with work. Last time I tested, null-move was worth maybe 100 Elo or so, when used in conjunction with LMR. LMR does a lot of what null-move does, and vice versa.

[Desperado]

Well, if you really talk of the "null-move-observation", there is
a clear NO. The null-move-observation is the fundamental
base of every quiescence search in todays standard search techniques.

In context to pruning and reduction techniques it may be possible to
reallocate the gain to other known algorithms like mentioned by Matthew
or like Bob said, an improved LM-technique may do the job.

Michael

[Barardo]

That's exactly what I was talking about.

What I really should have asked was: are there sufficient algorithms to allow one to create an engine that doesn't use The null move observation? What would it use for search and evaluation?

[hgm]

Well, if you really talk of the "null-move-observation", there is
a clear NO. The null-move-observation is the fundamental
base of every quiescence search in todays standard search techniques.

I don't think the null-move observation can be equated to stand-pat. It says there is almost always a better move than a null move. But stand-pat scores are not supposed to be lower bounds, and thus do not depend on that assumption. They are simply estimates of how good the position is. And they only have to be any good in positions where the best move is not a capture, as in QS captures are explicitly searched.

It is quite possible to design a branch-termination scheme that violates the null-move-observation, and would be more accurate. E.g. taking note of forks and skewers against the side to move, and assign scores below current eval when they exist. This is usually not competitive, because it is computationally expensive, but that is a different matter.

[Sven]

Well, if you really talk of the "null-move-observation", there is
a clear NO. The null-move-observation is the fundamental
base of every quiescence search in todays standard search techniques.

I don't think the null-move observation can be equated to stand-pat. It says there is almost always a better move than a null move. But stand-pat scores are not supposed to be lower bounds, and thus do not depend on that assumption. They are simply estimates of how good the position is. And they only have to be any good in positions where the best move is not a capture, as in QS captures are explicitly searched.

It is quite possible to design a branch-termination scheme that violates the null-move-observation, and would be more accurate. E.g. taking note of forks and skewers against the side to move, and assign scores below current eval when they exist. This is usually not competitive, because it is computationally expensive, but that is a different matter.

CPW supports more what Michael wrote, and I have the same opinion. Regarding the QS example, stand-pat scores are of course supposed to be lower bounds within the QS context. The name "null move observation" might be misleading since what is meant is not exactly the "null move" (as in null move pruning, do a virtual move that only changes the active color) but a "do nothing, stop searching at this node". In this sense, the stand-pat score in QS is used under the assumption that from all quiet moves there will be at least one that is at least as good as the estimation given by the static stand-pat score, since otherwise we could not use that score as a reference against which we compare all captures.

Sven

[Desperado]

Well, if you really talk of the "null-move-observation", there is
a clear NO. The null-move-observation is the fundamental
base of every quiescence search in todays standard search techniques.

I don't think the null-move observation can be equated to stand-pat. It says there is almost always a better move than a null move. But stand-pat scores are not supposed to be lower bounds, and thus do not depend on that assumption. They are simply estimates of how good the position is. And they only have to be any good in positions where the best move is not a capture, as in QS captures are explicitly searched.

It is quite possible to design a branch-termination scheme that violates the null-move-observation, and would be more accurate. E.g. taking note of forks and skewers against the side to move, and assign scores below current eval when they exist. This is usually not competitive, because it is computationally expensive, but that is a different matter.

Well i think you can suppose a standPat score to be a lower bound.

doNullmove(),
standPat = -evaluate();
undoNullmove()

if(standpat>=beta) return(standpat)

There it is, the NMO. We did a nullmove, and our argument is that if the nullmove fails high, there is at least another one which is going to be better than the nullmove. So, at the moment i cannot follow you.

The different matter isnt really a different matter in the context of the
original question. For me a 2800 elo Engine is a competitive one.

But you are right with the point, that a different branch termination scheme is possible. And maybe it is indeed possible from the viewpoint
of a super strong engine like Stockfish for example, to try out sth. in
that direction. It would be interesting if Stockfish with another branch
termination scheme could hold the level of 2800. A buffer of 400 Elo is a lot.

[bob]

Well, if you really talk of the "null-move-observation", there is
a clear NO. The null-move-observation is the fundamental
base of every quiescence search in todays standard search techniques.

In context to pruning and reduction techniques it may be possible to
reallocate the gain to other known algorithms like mentioned by Matthew
or like Bob said, an improved LM-technique may do the job.

Michael

I don't agree. The null-move observation is, simply "If I pass, and let my opponent make two moves in succession, and he STILL can not come up with a winning advantage, then his position is very bad, and mine is very good."

I don't see that concept in q-search. That's not the same as either standing pat, or trying a check or capture.

[bob]

Well, if you really talk of the "null-move-observation", there is
a clear NO. The null-move-observation is the fundamental
base of every quiescence search in todays standard search techniques.

I don't think the null-move observation can be equated to stand-pat. It says there is almost always a better move than a null move. But stand-pat scores are not supposed to be lower bounds, and thus do not depend on that assumption. They are simply estimates of how good the position is. And they only have to be any good in positions where the best move is not a capture, as in QS captures are explicitly searched.

It is quite possible to design a branch-termination scheme that violates the null-move-observation, and would be more accurate. E.g. taking note of forks and skewers against the side to move, and assign scores below current eval when they exist. This is usually not competitive, because it is computationally expensive, but that is a different matter.

Well i think you can suppose a standPat score to be a lower bound.

doNullmove(),
standPat = -evaluate();
undoNullmove()

if(standpat>=beta) return(standpat)

There it is, the NMO. We did a nullmove, and our argument is that if the nullmove fails high, there is at least another one which is going to be better than the nullmove. So, at the moment i cannot follow you.

Sorry, that is NOT null-move. The null-move-observation, as defined by Beal, was simply "If one side makes two consecutive moves without an intervening move by the opponent, and he can not produce a significant advantage by doing so, then his position is very bad, or, in our parlance, the side playing the null have a very strong position.

In the above, there is no "second consecutive move" so there is no "null-move-observation criterion that has been satisfied. You have to do a null-move and then let the opponent move again, else it is not a valid "null-move search."

The different matter isnt really a different matter in the context of the
original question. For me a 2800 elo Engine is a competitive one.

But you are right with the point, that a different branch termination scheme is possible. And maybe it is indeed possible from the viewpoint
of a super strong engine like Stockfish for example, to try out sth. in
that direction. It would be interesting if Stockfish with another branch
termination scheme could hold the level of 2800. A buffer of 400 Elo is a lot.

I already know the answer and have posted this previously. I ran 30K games using Crafty, with and without null-move. Maybe 120 Elo loss, max. The thread can be found with a search. Maybe 1-2 years ago max.