4

I am using the Arena GUI to run Stockfish. I set up the following position.

[FEN "r2q1rk1/2p1bppp/p1n5/1p1pP3/3Pn3/1P3N1P/1P3PP1/RNBQR1K1 w - - 0 11"]
[SetUp "1"]

At a search depth of 16 half moves stockfish gives the following analysis:

Nc3 f5 exf6 Nxf6 Bg5 h6 Qe2 Qd6 Qe6+ Kh8 Qxd6 cxd6 Bf4 b4 Na4 Nd7 Bg3 Rc8 Rc1 Kh7 Re6

As it stands, the line above consists of 21 half moves. I thought it was a problem with the GUI so I ran the same position into stockfish from the command line and again it gives a line with 21 moves. How is this possible? Is it a configuration from the engine to search at a depth higher than the specified?

Wais Kamal
  • 3,022
  • 1
  • 16
  • 41
  • 4
    Depth is obsolete measure. It's used only for historical reasons. It doesn't mean what you expect in modern engines. It's just a number with no exact meaning. – hoacin Nov 25 '17 at 17:40
  • So why even is the "number of half-moves" depth option there? – Wais Kamal Nov 25 '17 at 18:42
  • @WaisKamal is that the initial analysis or has it already played a few moves? – Dennis Nov 25 '17 at 18:49
  • 4
    People want to see easy number, they don't have to understand what's behind that. The bigger, the better. That's all. No reason to change this very old depth measure. It only doesn't mean what it was named after some decades ago. – hoacin Nov 25 '17 at 21:14
  • No @Dennis I restarted the engine to clear the hash memory then analysed this position. – Wais Kamal Nov 26 '17 at 15:58

1 Answers1

7

Computers are really stupid, they can't evaluate a position that contains exchange threats. When the program reaches the expected depth and there are exchanges present in the position, the computer performs a Static Exchange Evaluation, which is hard to program correctly, or it extends the search depth until the exchanges and other factors are eliminated.

Fred Knight
  • 3,112
  • 10
  • 8
  • 1
    That sounds more like computers are clever enough not to evaluate highly dynamic positions. I don't see much stupidity in that. – hoacin Nov 27 '17 at 08:37
  • @hoacin I'd argue that cleverness should be attributed to the programmers that told them to not evaluate immediately. – Inertial Ignorance Aug 20 '18 at 10:58
  • Even 1980s chess programs like Sargon II would push beyond the normal number of ply for the set level in cases involving captures. – supercat Feb 24 '20 at 22:16
  • I think that the UCI output from info command will give a parameter called "seldepth". That should be the maximal search depth effective as per that extension from specified depth. googling gives: http://www.talkchess.com/forum3/viewtopic.php?t=65516 – dbdb Aug 07 '21 at 01:17