Yet Another Mate Solving Test

[Paloma]

Yes, after 1. Na3 Kb6 needs X44.
Ignacio.

after 1. Na3 Kb6 i get mate in 8 instead mate in 9 ?!?

Suche nach Spezial-Matt [C0/R0/K4/P10/X44] in 9 ... (Hash=256MB)
d=1 t=0.00s [1.00] n=0kN [1.00] Hash: spd=1.00 in=1 out=0
d=2 t=0.00s [1.00] n=0kN [1.00] Hash: spd=0.98 in=1 out=0
d=3 t=0.00s [1.00] n=1kN [12.18] Hash: spd=0.98 in=39 out=0
d=4 t=0.01s [1.00] n=8kN [12.40] Hash: spd=1.10 in=472 out=0
d=5 t=0.20s [20.00] n=125kN [14.98] Hash: spd=1.23 in=5067 out=0
d=6 t=4.17s [20.85] n=2663kN [21.39] Hash: spd=1.50 in=110805 out=0
d=7 t=64.68s [15.51] n=42026kN [15.78] Hash: spd=1.91 in=1769615 out=96
8.20 9:05 +M8 2.Sxb5 707
d=8 t=1103.62s [17.06] n=757264kN [18.02] Hash: spd=2.36 in=29330885 out=22620022
Suche abgeschlossen ... (Zeit=1103.62s)
Matt in 8 gefunden ! (1 Lösung in 18:23)
8/08 18:23 +M8 2.Sxb5 cxd6 3.Sc7+ Ka7 4.Sxd5 Thc8 5.Db6+ Ka8 6.Sd4 h6 7.Sb5 axb5 8.Da5+ La6 9.Dxa6+ 686
Bester Zug: Sa3xb5 Zeit: 18:24.051 min K/s: 686.164 Knoten: 757.264.321

[Ignacio]

Yes Herbert, but the mate in 8 requers X44.
BTW: on my PC Chest needs 1 second for the mate in 8 with parameters.
1r5r/1bp1Rppp/pk1B4/1p1q4/1Q6/N1P2N2/PP3PPP/R5K1 w - - bm #8; 00:01 @ C0/R2/K4/P10/X44;

Best.
Ignacio.

[Houdini]

Jouni,

One idea: add MATERIAL only mode to Houdini/Critter/etc for faster mate finding. There is one in Goliath engine.

I'm not convinced that using material eval only will speed up the mate finding. The efficiency of the search algorithm would probably be reduced if only material factors were taken into account.

I've toyed with another, probably more useful idea: take into account the information about the expected mate length, and limit the search tree (more or less) to the specified mate length. I've tested this with a new "MateSearch" UCI option in Houdini in which the maximum length of the mate can be specified.
Running your 100 test positions with 10 sec/position using 3 cores of a Core i5-750 gives the following results:

1) with the standard Houdini 1.5, I obtain 57/100 of which 31 require less than 1 second.
2) with the dev version using MateSearch=11, I obtain 63/100 of which 44 require less than 1 second.

Quite a nice improvement.

Robert