Very Hard Materialism Test Position for Dedicated Comps

[RadioSmall]

In the following position both the RadioShack 2250 XL and Novag Sapphire want to play 1.Qxb4?? which loses eventually (the variations are hard to see for both humans and dedicated Comps ) after a full 24 HOURS of Computation:
[fen]3r2k1/5pp1/7p/4P3/Qp1Pq3/6P1/P6P/3R2K1 [/fen]

The only dedicated Comp I tested that can avoid this prior to reaching a full 24 hours of thinking time is the Designer 2265 ! which selects the correct 1.Qb3! instead. To get an appreciation of How Hard this task is please note that Hiarcs 13.1 on my Palm Tungsten C (400 Mhz) still goes for 1.Qxb4?? after completing a full 18 ply search! .Hence Kudos to the Fidelity Designer 2265!!

[Steve B]

Hi RS

how long did it take the D2265 to hit upon 1.Qb3?
did it want to play this from the getgo or switched to it after several hrs?
what move/moves did it consider first?

actually i am not so sure what significance to place upon the fact that a dedicated computer stumbles upon the proper move after pondering for hrs and hrs .. unless you want to establish that this computer is good for analysis purposes .. however i doubt anyone uses dedicated computers for serious analysis these days given the plethora of PC engines

still.. the position is interesting

[fen]3r2k1/5pp1/7p/4P3/Qp1Pq3/6P1/P6P/3R2K1 w - - 0 71 [/fen]

i fed the position to the Montreal 68000 and the Roma ll..both programmed by Lang and both rated close to the D2265
initially each wanted to play 1.Qa7 then after about 30 seconds each chose 1.Qxb4 and stayed with this move for a full 60 minutes scoring the position as -.24

Separate But Equal Regards
Steve

[RadioSmall]

Hi RS

how long did it take the D2265 to hit upon 1.Qb3?
did it want to play this from the getgo or switched to it after several hrs?
what move/moves did it consider first?

actually i am not so sure what significance to place upon the fact that a dedicated computer stumbles upon the proper move after pondering for hrs and hrs .. unless you want to establish that this computer is good for analysis purposes .. however i doubt anyone uses dedicated computers for serious analysis these days given the plethora of PC engines

still.. the position is interesting

[fen]3r2k1/5pp1/7p/4P3/Qp1Pq3/6P1/P6P/3R2K1 w - - 0 71 [/fen]

i fed the position to the Montreal 68000 and the Roma ll..both programmed by Lang and both rated close to the D2265
initially each wanted to play 1.Qa7 then after about 30 seconds each chose 1.Qxb4 and stayed with this move for a full 60 minutes scoring the position as -.24

Separate But Equal Regards
Steve

Hi Steve
Always good to hear from you. Well, this is what happens with the Designer 2265 : It wants to play 1.Qa7 up to 11 seconds after which it changes to 1.Qb5 and stays with it till 41 seconds , when it changes to the blunder 1.Qxb4 . However at 51 seconds it changes to 1.d5 and stays with it for a long time up until 37 minutes and 5 seconds after which it switches to the blunder 1.Qxb4 . At 1 hour and 6 minutes it lets go of this blunder and wants to play 1.Rf1 .Finally it changes to 1.Qb3 at 1 hour and 41 minutes and does not change for the Duration of the thinking time. The important thing here is that non of the other moves considered by the Designer 2265 ( besides 1.Qxb4) are blunders and that at tournament time controls the D 2265 would have avoided this blunder and played 1.d5. This is very impressive for an eighties chess computer , I would say almost unbelievable. Since you mentioned Lang programs , it is also worth mentioning that on my Pocket PC (624 Mhz) Chess Genius for windows mobile needs 16 minutes and 39 seconds to avoid playing 1.Qxb4. On the Palm Tungsten C (400 Mhz ) Chess Genius needs 25 minutes and 53 seconds.
Counting minutes and seconds Regards
RadioSmall

[Steve B]

However at 51 seconds it changes to 1.d5 and stays with it for a long time up until 37 minutes and 5 seconds after which it switches to the blunder 1.Qxb4 . At 1 hour and 6 minutes it lets go of this blunder and wants to play 1.Rf1 .Finally it changes to 1.Qb3 at 1 hour and 41 minutes and does not change for the Duration of the thinking time. The important thing here is that non of the other moves considered by the Designer 2265 ( besides 1.Qxb4) are blunders and that at tournament time controls the D 2265 would have avoided this blunder and played 1.d5. This is very impressive for an eighties chess computer , RadioSmall

I Would have to agree with you then
thanks for the detailed report on the D2265's analysis

which dedicated chess computer programmers have we missed in testing this position?
i think Schroeder?..any others?
ill have Schroeder's MMV take a stab at the position
will be interesting to see if we can find another engine rated about the same as the D2265 that plays something other 1.Qxb4 within the first 60 minutes


Well Done Regards
Steve

[Steve B]

Interestingly
the MMV considered practically every move
EXCEPT 1.Qxb4

initially it would have played
1.Kf2 and then 1.h3

at the 2 minute mark it switches to 1.Qb3 and would have played it at the 3 minute mark
however in all fairness i must report that 10 seconds later it switches to 1.d5 and stays with this move for about 15 minutes
when it switches to 1.Qb5..(carefully avoiding the poisoned pawn)
it stays with 1.Qb5 for the remaining 45 minutes scoring the position -.09

Close Regards
Steve

[Cubeman]

Good test position, to make sure that comps that avoid 1.Qxb4 really see the refutation you need to test them on the continuation 1..Qe3+ 2.Kf1 (forced) 2..Qf3+ 3.Ke1 and then the real test is to see that they play the winner 3..Rc8!! some will play the 3..Qh1+ and 4..Qxh2+ then 5..Qxg3 which some think is better for black.
[fen]3r2k1/5pp1/7p/4P3/1Q1P4/5qP1/P6P/3RK3[/fen]w
Test how long to find 3..Rc8!!

[RadioSmall]

Interestingly
the MMV considered practically every move
EXCEPT 1.Qxb4

initially it would have played
1.Kf2 and then 1.h3

at the 2 minute mark it switches to 1.Qb3 and would have played it at the 3 minute mark
however in all fairness i must report that 10 seconds later it switches to 1.d5 and stays with this move for about 15 minutes
when it switches to 1.Qb5..(carefully avoiding the poisoned pawn)
it stays with 1.Qb5 for the remaining 45 minutes scoring the position -.09

Close Regards
Steve

Thanks for the report on the MMV .The calculations here are also very impressive. These days people in general are so busy with PC software that the true value of the dedicateds goes unnoticed. It looks like with an hour per move on a good pre 1995 dedicated machine one gets very accurate analysis in most chess positions. I have also been lately curious about the (theoretical ) strength of dedicated computers at very long think times like 24 hours per move. How much elo would a good dedicated Computer gain if it could make the same moves it makes say at 24 hours per move at tournament time controls??
Wondering Regards
RS

[deni14]

I have been excited by this example, for the simple machines seem to play it better! Both shredder (ipad) and another similar program could not play Qb3! (in over an hour) But the old fidelity dedicateds play it easily. The S9 plays Qb3 in 3 minutes, the design 2000 5 minutes, and the design (display) 2100 (and Par) in 2.5 minutes. A tasc 2.5 takes 16 minutes, and a Genius skips over it for over 90 minutes before we stopped it!

What does this mean then ? It seems to indicate that the newer expensive chess computers may not always play as well as an Sensory 9 ! Anyway this will keep me busy awhile trying to figure out what is going on. Appreciate any suggestions why this is happening.

Dennis

[Cubeman]

Don't test the original position, as some comps will discard 1.Qxb4 not because of the refutation on move 3 but because of allowing the black Queen to win the h2 and possibly g3 pawn. The real test to see if comps really understand why 1.Qxb4 is a mistake is to retest them on the position I posted earlier. I will wait for results from Radio Small and hopefully Steve B.

[Steve B]

Don't test the original position, as some comps will discard 1.Qxb4 not because of the refutation on move 3 but because of allowing the black Queen to win the h2 and possibly g3 pawn. The real test to see if comps really understand why 1.Qxb4 is a mistake is to retest them on the position I posted earlier. I will wait for results from Radio Small and hopefully Steve B.

As the MMV was the only computer i tested which did not swallow the poisoned pawn i gave it the position after Whites 3.Kf1
[fen]3r2k1/5pp1/7p/4P3/1Q1P4/5qP1/P6P/3RK3 w - - 0 1[/fen]

it immediately wants to play 3..Qh1+ and sticks with it for the full 60 minutes ..scoring the position +.86

Oh Well Regards
Steve

[Cubeman]

Thanks for the reply, The Qh1+ line also seems to lead to an advantage but not as clear cut as the Rc8!! line. So at the end of the day dedicated should still refrain from 1.Qxb4 as they should at least see some sort of advantage to Black. The trouble comps have with the original position is that White is in a bad way and has to lose at least a pawn and the best way is as the Designer showed with 1.Qb3 as if white decides to win the pawn then it will peter out to a probable draw with just Queens on the board.I might try and test Mephisto Expert Travel which I think is a Morsch program.

[Cubeman]

In the following position both the RadioShack 2250 XL and Novag Sapphire want to play 1.Qxb4?? which loses eventually (the variations are hard to see for both humans and dedicated Comps ) after a full 24 HOURS of Computation:
[fen]3r2k1/5pp1/7p/4P3/Qp1Pq3/6P1/P6P/3R2K1 [/fen]

To get an appreciation of How Hard this task is please note that Hiarcs 13.1 on my Palm Tungsten C (400 Mhz) still goes for 1.Qxb4?? after completing a full 18 ply search! .Hence Kudos to the Fidelity Designer 2265!!

Hiarcs 13.1 running in PocketFritz4 on my Acer Neotouch (1Ghz) rejects 1.Qxb4 at depth 15 and sees the line in it's PV 1.Qxb4 Qe3+ 2.Kf1 Qf3+ 3.Ke1 Rc8 4.e6 (-4.00) Depth 15 00:50 and then switches to 1.Rf1 Qxd4+ 2.Rf2 Qc4 3.Kg2 Qe4+_ Kh3 (-0.53) Depth 16 01:42
The versions of Hiarcs 13.1 behave differently on Palm and Windows Mobile.

[Cubeman]

Thanks for the reply, The Qh1+ line also seems to lead to an advantage but not as clear cut as the Rc8!! line. So at the end of the day dedicated should still refrain from 1.Qxb4 as they should at least see some sort of advantage to Black. The trouble comps have with the original position is that White is in a bad way and has to lose at least a pawn and the best way is as the Designer showed with 1.Qb3 as if white decides to win the pawn then it will peter out to a probable draw with just Queens on the board.I might try and test Mephisto Expert Travel which I think is a Morsch program.

The Expert Travel also fails both tests, falls for 1.Qxb4 and 3..Qh1+

[Steve B]

I tested both positions with Ulf Rathsman's Mephisto B&P module which is only rated 1716 by Sel.Search Mag.

in the first position it will not take the poisoned P and plays 1.Qb5

More Surprisingly ..in Cubeman's position...
it will play 3..Rc8!

both moves are played within 60 minutes

Startled Regards
Steve

[RadioSmall]

Good test position, to make sure that comps that avoid 1.Qxb4 really see the refutation you need to test them on the continuation 1..Qe3+ 2.Kf1 (forced) 2..Qf3+ 3.Ke1 and then the real test is to see that they play the winner 3..Rc8!! some will play the 3..Qh1+ and 4..Qxh2+ then 5..Qxg3 which some think is better for black.
[fen]3r2k1/5pp1/7p/4P3/1Q1P4/5qP1/P6P/3RK3[/fen]w
Test how long to find 3..Rc8!!

Hi
The Novag Sapphire finds 3...Rc8!! in 5 minutes and 15 seconds at depth 8. So this new test position is much easier ( at least for the Sapphire ) .I will test some other computers also.