interesting "null move killer" position

[jdart]

A position with a fair amount of material where null move can malfunction:

[d] 2Q2bk1/5pp1/p3p2p/q1p1P2P/1pP5/1P2B1P1/P4PK1/8 w - - 0 1

bm is Qb8. This actually occurred in an offline game. Arasan 10.1 does ok on this but my latest build has trouble, for some reason.

--Jon

[rightrook]

Ruffian 1.0.5 plays g4.....

but my Junior 5.0 finds Qb8 in just 6 seconds...!



Robert

[BubbaTough]

LearningLemming0.24 finds Qb8 in 1 second, so take that Junior. Not so tough now that you aren't running on 16 processors like you were when you crushed me in CC10 are you! .

Assumably, any engine doing any form of null move verification will handle this position fine, those that aren't won't. Of course, the question always is, does this come up enough to be worth it (or, as in my case, is it so annoying to you to not find the right move in these cases that you want to verify even if it ends up making your engine slightly weaker (as measured by rating performance anyway).

-Sam

[Jose Carlos]

Are you sure this position is a null move killer? Averno gets it instantly, and Averno is anything but good in null move killer positions.

[bob]

LearningLemming0.24 finds Qb8 in 1 second, so take that Junior. Not so tough now that you aren't running on 16 processors like you were when you crushed me in CC10 are you! .

Assumably, any engine doing any form of null move verification will handle this position fine, those that aren't won't. Of course, the question always is, does this come up enough to be worth it (or, as in my case, is it so annoying to you to not find the right move in these cases that you want to verify even if it ends up making your engine slightly weaker (as measured by rating performance anyway).

-Sam

I don't do any sort of verification whatsoever, and get this:

Code: Select all

               16->   3.16   1.25   1. Kf3 Qb6 2. Ke4 Qa5 3. Kf4 Qb6 4.
                                    Kf3 Qa5 5. Kg4 <HT>
               17     7.33   1.19   1. Kf3 Qb6 2. Ke4 Qa5 3. Kf4 Qb6 4.
                                    Kf3 Qa7 5. Qc6 a5 6. Ke4 Be7 7. Qe8+
                                    Bf8 8. Qd8 Qb7+ 9. Kd3 Qa7 10. Ke4
               17    10.04     +1   1. Qb8!!
               17    12.41     +3   1. Qb8!!
               17    16.90   3.17   1. Qb8 Qxa2 2. Bxc5 Qxb3 3. Qxf8+ Kh7
                                    4. Be7 Qxc4 5. Qxf7 Qe4+ 6. Kh2 Qxe5
                                    7. Bxb4 a5 8. Bd2 Kh8 9. Qe8+ Kh7 10.
                                    Qg6+ Kh8 11. Bf4 Qd5 <HT>

That on my rather old dual xeon 2.8 in my office...

My laptop found it in under 3 seconds...

[jdart]

It is related to null move, I think, because Black is in zugzwang after Qb8. But some engines want to play g4.

[YL84]

Hi,
my engine Milady has no null move (just LMR) and thus seems not to have difficulties here. Qb8 is found in 0.48s... Below is the analysis. Thanks for the position !
Yves

"depth 1 score 30 temps 0 nbnod 33 B_F :32 : g2g1"
"depth 2 score 40 temps 1 nbnod 122 B_F :9,4339811320566 : g2g1 a5a2"
"depth 3 score 20 temps 5 nbnod 722 B_F :8,43432665301749 : c8e8 a5b6 a2a4"
"depth 4 score 40 temps 12 nbnod 1941 B_F :5,90882225999329 : c8e8 a5b6 e8d7 f7f5"
"depth 5 score 80 temps 48 nbnod 9046 B_F :5,89267944278086 : c8b8 f7f5 e5f6 a5a2 f6g7"
"depth 6 score 130 temps 119 nbnod 19742 B_F :4,69393313773888 : c8b8 f7f5 e5f6 g8f7 f6g7 f7g7"
"depth 7 score 160 temps 251 nbnod 48241 B_F :4,32916592676341 : c8b8 f7f5 e5f6 g8f7 f6g7 f7g7 b8e5"
"depth 8 score 160 temps 506 nbnod 91544 B_F :3,79808559674428 : c8b8 f7f5 e5f6 g8f7 f6g7 f7g7 b8e5 g7f7"
"depth 9 score 170 temps 1114 nbnod 224367 B_F :3,70896373738002 : c8b8 f7f5 e5f6 g8f7 f6g7 f7g7 b8e8 a5a2 e3h6"
"depth 10 score 170 temps 2370 nbnod 439983 B_F :3,41482221854569 : c8b8 f7f5 e5f6 g8f7 f6g7 f7g7 b8e8 a5a2 e3h6 g7h6"
"depth 11 score 220 temps 5104 nbnod 925792 B_F :3,28814843335134 : c8b8 f7f5 e5f6 g8f7 f6g7 f8g7 b8a7 f7g8 e3c5 a5a2 a7b8"
"depth 12 score 220 temps 10548 nbnod 1776995 B_F :3,12010657585732 : c8b8 a5a2 e3c5 a2b3 b8f8 g8h7 c5b4 b3c4 f8d6 h7g8 d6d8 g8h7"
"depth 13 score 220 temps 18814 nbnod 3316745 B_F :2,99197326592976 : c8b8 a5a2 e3c5 a2b3 b8f8 g8h7 c5b4 b3c4 b4d6 c4d5 f2f3 d5d2 g2g1"
"depth 14 score 200 temps 33927 nbnod 5582504 B_F :2,84409328329356 : c8b8 a5a2 e3c5 a2b3 b8f8 g8h7 c5b4 b3c4 b4d6 c4d5 f2f3 d5d2 g2h1 f7f6"
"depth 15 score 200 temps 55752 nbnod 9775268 B_F :2,76375809640694 : c8b8 a5a2 e3c5 a2b3 b8f8 g8h7 c5b4 b3c4 b4d6 c4d5 f2f3 d5d2 g2h1 f7f6 e5f6"

[BubbaTough]

My laptop found it in under 3 seconds...

That is an interesting result bob. I tried a few second search without verification and LearningLemming likes g4. I wonder if Crafty is doing something subtly different (as does Junior)? With verification this is maybe a 5 ply combo with lots of captures, so the fact that you and junior are finding it quite a bit later than a verification search does (17 ply for you, maybe more for Junior if it is taking 6 whole seconds) makes me suspicious that perhaps the programs are just stumbling on to the solutions. It might be interesting to have a testbed of simple zugzwang positions with significant material on (its not really that odd to come up in games where humans are squeezing a large space advantage) and see how reliable you non-verifier searches are at solving these. Unless you already know a reason you can reliably find it without verification?

-Sam

[Uri Blass]

Are you sure this position is a null move killer? Averno gets it instantly, and Averno is anything but good in null move killer positions.

I think that it is dependent on the program.

Qb8 is a move that may have threat or not have threat dependent on the evaluation.

If it has a threat you do not prune it so you find it regardless of verification.
If it has no threat then you prune it and you never find it if you do not do verfication.

Uri

[pedrox]

DanaSah with verification of null move sees Qb8 at 0 seconds, but without verification prefers g4.

AICE play g4