Mates that are difficult for engines

Here's a White to move and mate in 69:

White to move and mate in 64, from 1879:

Here is one that I've been working on for a while to determine the shortest mate. White to move and mate in 11 moves:

Here is a very tough one, White to move and mate in 13:

@drdos7 @Arisktotle

To continue my "anal retentiveness" (because I'm not overly keen on the results when it's indiscriminately avoided); after some investigation I've reached the conclusion that SF doesn't lie with it's mate announcements. Rather the GUIs that most people use to display them ubiquitously do. On the other hand these don't normally fail to render a mate announcement if SF makes one, so if the absence of such is taken to mean SF has difficulty in finding a mate, that is probably reliable. The contrapositive conclusion that a mate announcement rendered by the offending GUIs means SF does not have difficulty finding a mate is, however, not reliable.

Well then Martin, the solution is to run Stockfish in terminal mode if you are that anal retentive thus bypassing the GUI if you believe it is the case that the GUI is lying or misrepresenting Stockfish's output.

I love that one because it looks also extremely hard for humans. Usually it's always an inverse relationship between them, the ones hard for humans, like those hard mate in 8s are easy for engines, but those "easy" mate in 67s that are simple concepts for humans to understand, always are hard for most computers. I love the ones that are hard for both!

The other really nice thing about that one is that White sacrifices all of his pieces except for a single pawn that delivers the mate!

Is it the only way to win though, since white is up so much material and black doesn't have any immediate threats?

No, it is NOT the only way to WIN, but you are in my thread here which stipulates a mate in 13, not White to play and win, however Black CAN stop White from sacrificing all of his pieces by playing different moves, but it still ends up being a mate in 13, here's an example, with an alternate line after BLACK varies on move 9:

Young man, that one would be good except that Black can force a draw by playing 2...Rc6-c5 instead of 2...Rc4-c5 and of course at other points in the solution you gave.

If white just tried to win without seeing that mate in 13 line, how many moves to an obvious win/slightly longer mate would it be?

The position is a win for White from the very start, so no moves needed just to get a win, however if you wanted an alternate longer mate then 1.Bg4 will give you a mate in 15 in the following manner:

Indeed, I came to the above conclusions by checking the info returned in terminal mode against the Arena output for the KNNKP position I posted earlier. SF15.1-NNUE's search appears to be completely reproducible (neglecting the timestamps) for this position (though curiously not for the KBNK position I posted).

Arena doesn't actually lie. If you look at the Arena output I posted earlier, each apparent mate announcement where a later valuation appears to be worse is followed by a "+". These correspond in that case to info returned by SF containing the "lowerbound" keyword. I would speculate that means SF has determined there is no better valuation starting with the moves that follow. These are not valuations. I say "speculate" because there appears to be no user or developer documentation to that effect for SF. (The source is open code, so if users wish to navigate GitHub's IDE and learn any of the associated computer languages they'e not already familiar with, you could take the view that the source represents complete documentation.)

Unfortunately Arena emulates the Baker in The Hunting of the Snark by omitting to mention the fact. All the Arena user documentation says about the info lines is "The next lines contain the current mainlines (principal variation) with depth, time, nodes, nodes per second, and value". The first field is obviously not depth in any of the lines and there is no mention that some of the lines may be boojums.

Nevertheless running SF in terminal mode is not the only solution. Arena is adequate so long as you understand the significance of the "+" and "-" signs that appear. There have been several threads asking what the symbols mean in the forum, but no correct answers that I've seen (apart from one I just posted myself).

Unfortunately Arena is the only GUI I've come across where that applies. The rest appear to display the upperbound and lowerbound returns as normal valuations without any indication that they're not.

The relevance to the thread is the interpretation of "engines find difficult". A mate displayed in a boojum is not an indication that the engine is not finding it difficult, but the majority of people probably take it as so. This would no doubt be the case with any engine, not just SF. The root cause is, I would say, a lack of anal retentiveness on the part of the people specifying the UCI protocol.

Edit: 

I just realised that the hypotheses I posted above are far too optimistic. The Arena/SF KBNK example I posted earlier has out of order mate announcements with no "+" signs. (I didn't manage to reproduce it from the command line; that produced only one lowerbound which wasn't a mate announcement and the mate announcements were in descending depth.)

That suggests that SF itself does "lie" about the mates and driving it from the command line or in Arena won't necessarily help when it does.

An alternative hypothesis (your guess is as good as mine) is that the mate announcements are valid only for the moves shown and generally don't represent a fully searched mate at all.

I think it remains true that you can conclude that if, from a position where there is a forced mate, no mate announcements are displayed in an hour or so then SF or whatever engine is having difficulty with it.

Arena now doesn't work cause it claims a DLL file is now missing even though it worked the first time. Why isn't there just a stockfish app for windows like there is with phones, and why doesn't the stockfish engine run on chessbase? Absolutely insane.

Missing DLL sounds like something you managed yourself rather than Arena's fault. Try reinstalling.

SF developers probably have no interest in writing a GUI. More interested in beating other engines. MS probably even less interested. Plenty of apps available mostly free.

Chessbase looks commercial, so lack of SF could be connected with the GNU license (or not - I don't look into such things, I just want them free).

Here's a mate in 17 that is tough, most engines will find a mate but probably not a mate in 17 which is the stipulation for this white to move and mate in 17:

Thirty mates that Stockfish (without tablebase) finds difficult. (It would by easy, if tedious, to post 30,000 similar). All White to play.

Thanks, I'm going to try all of them out sans tablebases of course.

Sixty more mates that Stockfish (without tablebase) finds difficult - but I'm not sure there are any humans (without tablebase) about that don't also find these difficult. Again all White to play.