Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Rook Sacrifice and a deadly fail low

Author: Uri Blass

Date: 11:35:08 03/07/04

Go up one level in this thread


On March 07, 2004 at 13:46:41, Anthony Cozzie wrote:

>On March 06, 2004 at 17:06:23, GeoffW wrote:
>
>>Hi
>>
>>Here is a position I found interesting, white has just played Rf6, black to move
>>
>>[D]r3qnk1/rb2bppp/4pR2/pp1pP1PP/2pP3N/P1P1P1Q1/1PB5/4B1RK b - - 0 1
>>
>>My programs analysis for this position is horrid
>>
>>Ply  Eval   Time      Nodes   Principle Variation
>> 1    433      0         17   g7f6 e5f6
>> 2    433      0        290   g7f6 e5f6
>> 3-   393      1       4782   g7f6
>> 3    204      1       8053   e7a3 b2a3 g7f6 e5f6
>> 4    204      3      18575   e7a3 b2a3 g7f6 e5f6
>> 5    204      6      44876   e7f6 g5f6 g7g6 h4f3 f8d7 h5g6 h7g6
>> 6    192     10      97670   e7f6 e5f6 g7g6 h5g6 h7g6 h4f3
>> 7    194     21     193103   e7f6 e5f6 g7g6 h4f3 a8d8 h5g6 h7g6
>> 8    187     60     564304   e7f6 e5f6 g7g6 h4f3 g6h5 g3h4 f8d7 h4h5
>> 9    188    132    1263805   e7f6 e5f6 g7g6 h4f3 f8d7 e1d2 a8b8 g3h2 e8d8
>>10    176    396    3783030   e7f6 e5f6 g7g6 h4f3 f8d7 g3h2 a8d8 e1g3 d7b6 f3e5
>>11    176    942    8062241   e7f6 e5f6 g7g6 h4f3 e8c6 g3h2 f8d7 e1g3 a8d8 f3e5
>>d7e5 g3e5
>>12-   136   3984   35236639   e7f6
>>12    102  144954  868931003   b5b4 a3b4 a5b4 f6f4 a7a2 g3g2 b4b3 c2b1 a2a1 e1d2
>>f8d7 e3e4
>>
>>It firstly thinks it has won a rook by taking it with Pxf6, then it realises
>>this is bad, and avoids the Rook sacrifice by coming up with its own weird
>>bishop sacrifice Bxa3. Two plies later then decides it had better take the rook
>>after all with the obvious looking Bxf6. It takes 40 seconds to get to ply 12
>>before it realises that this is also a hopeless losing move.
>>Then it takes almost 25 minutes to resolve the fail low with a branching factor
>>of 24 !!
>>I am not sure why it struggles to resolve this specific fail low so badly, this
>>is probably the trickiest fail low I have seen so far, at least for my program.
>>It was taking so long I was convinced it was locking up in an infinite search
>>bug. I guess it probably isnt a bug as it finally does come up with a reasonable
>>move, all be it far too late.
>>
>>What do other programs make of this, daresay Crafty et al. will have it all
>>worked out and resolved in about 5 seconds flat ;-(
>>
>>P.S it finally managed ply 13 as I was typing this up
>>
>>13    102  327723 2072998497   b5b4 a3b4 a5b4 f6f2 b4b3 c2b1 f8d7 e1d2 a7a1 h5h6
>>g7g6 h4f3 a8a7
>>
>>54 minutes to decide white is a pawn up after avoiding the sac.
>>
>>         Regards Geoff
>
>
>Working out fail lows always takes a long time :(
>
>Zappa needs 1.4 seconds (opteron 1.8) to avoid Be7xf6 (gxf6 is only played by
>the q-search ;) I think black is in some trouble here though:
>
>1... Be7xf6 2. g5xf6 g7-g6 3. Qg3-g5 Nf8-d7 4. h5xg6 h7xg6
> = (0.90)       Depth: 7/10     00:00:00.13     62650kN
>1 ...Be7xf6--
> - (0.65)       Depth: 8/9      00:00:00.14     64878kN
>1 ...Be7xf6--
> - (0.65)       Depth: 8/9      00:00:00.38     163136kN
>1... Be7xf6 2. g5xf6 g7-g6 3. Qg3-g5 Nf8-d7 4. h5xg6 h7xg6 5. Nh4xg6
> = (-0.86)      Depth: 8/12     00:00:00.40     176097kN
>1... Bb7-c6 2. Rf6-f2 Qe8-d8 3. h5-h6 g7-g6 4. Nh4-f3 Nf8-d7 5. Rg1-f1
> = (-0.54)      Depth: 8/14     00:00:01.31     592471kN
>1... b5-b4 2. a3xb4 a5xb4 3. Rf6-f1 b4-b3 4. Bc2-b1 Ra7-a1 5. Be1-d2
> = (0.26)       Depth: 8/14     00:00:03.07     1503664kN
>1... b5-b4 2. c3xb4 a5xb4 3. Be1xb4 Be7xb4 4. a3xb4 Bb7-c6 5. Rf6-f1 Qe8-e7 6.
>g5-g6 f7xg6 7. h5xg6 h7-h6 8. Qg3-e1
> = (-0.69)      Depth: 14/25    00:04:59.46     217960622kN
>
>(score = white up .7)
>
>Top programs will get this one with eval I think.
>
>anthony

I think that you overestimate the top programs.
I think that the search is the main advantage of them.

I guess that all of them see advantage for black at small depth and suggest Bxf6
at small depth.

Uri



This page took 0.01 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.