Critter doing really deep search

[Jouni]

From Rybka forum

[D]2r1k3/2r5/P2Q1p2/b5Bp/2P2P1P/6P1/5K2/7b w - - 0 1

Analysis by Critter 1.0 64-bit SSE4:

...
1.Qe6+ Re7 2.Qxc8+ Kf7
+- (78.70) Depth: 27/119 03:06:41 141823mN, tb=7569141
1.Qe6+ Re7 2.Qxc8+ Kf7
+- (78.70) Depth: 28/119 03:15:05 147318mN, tb=8968203
1.Qe6+ Re7
+- (82.85) Depth: 29/119 04:12:23 188146mN, tb=12998707
1.Qe6+ Re7
+- (95.24) Depth: 30/140 06:54:33 306952mN, tb=42748670
1.Qe6+ Re7
+- (#85) Depth: 31/190 23:31:21 1131025mN, tb=345803194

Is 190 plies new record for playing engines!?

Jouni

[bob]

Where do you get 190 plies? Most likely its an EGTB hit. And we have seen far deeper mates announced than mate in 85.

[Jouni]

"31/190 " means 190 plies selective as UCI engines display

Jouni

[marcelk]

"31/190 " means 190 plies selective as UCI engines display :)

Jouni

I'm curious, do people really pay attention to that 'selective depth' number?
It was introduced once in engines as some kind of marketing gimmick.
But from a programmer's point of view you want it to be as low as possible.

[Dann Corbit]

"31/190 " means 190 plies selective as UCI engines display

Jouni

I'm curious, do people really pay attention to that 'selective depth' number?
It was introduced once in engines as some kind of marketing gimmick.
But from a programmer's point of view you want it to be as low as possible.

I think in general it just means search depth + quiescent length to finish
so it is not really very helpful most of the time.

At least one engine uses the selective depth field simply to tell you how long the displayed pv was in nodes

[Sven]

It is very impressive, of course.

In the given position, after playing around with it for a while I would expect that it is a forced mate in less than 60 plies. After winning the Rc8 white will also remove one of the two black bishops from the board, and very soon the white passers a6/c4 will cost more black material.

Does anyone see how black could survive longer than about 55-60 plies here, until being mated?

For instance, after 1. Qe6+ Re7 2. Qxc8+ Kf7 3. Bxf6 Be1+ 4. Kg1 Kxf6 5. Kxh1 Bf2 6. Qh8+ Kf7 7. c5 black cannot avoid losing both rook and bishop within few moves, leaving white with a queen and some pawns against the bare king. Other variations look similar.

It is clear that a forced mate in, say, 60 plies will usually not be found in iteration no. 31. But here Critter obviously found some very deep lines, so I wonder why Critter did not display something like "#30", provided my assumption above is correct.

A PV longer than two or four plies would have been interesting in this case ...

Sven

[Jouni]

Houdini thinks this is mate in 18-20 moves. Critter is really doing too much extensions and only reached 31 plies in 23(!) hours.

Jouni

[mhalstern]

Nobody on the Rybka forum answered the poster's question:

Why didn't Critter show the complete line leading to the announced mate?