The New Sudoku Players' Forum

[denis_berthier]

How would you react if someone announced (without giving the puzzle) (s)he has found a 9x9 single-solution Sudoku puzzle with SER 12.0 (or more)? Why?

(N.B.: I'm not suggesting I've found one (I've never searched for hard puzzles). I just want to know how you'd react).

[Edited for precision]

[Leren]

Perhaps I would react by saying it has already been done, see eg here. Also, I would wonder whether eleven would change his moniker to twelve .

Leren

[denis_berthier]

I mean a real Sudoku puzzle (not a Sukaku), with a single solution.

[m_b_metcalf]

I mean a real Sudoku puzzle (not a Sukaku), with a single solution.

So my x-sudokus also wouldn't count?

Mike

[denis_berthier]

I mean a real Sudoku puzzle (not a Sukaku), with a single solution.

So my x-sudokus also wouldn't count?
Mike

Hi Mike,
I was not aware of this puzzle.
It's certainly a very interesting example for X-Sudoku and I'm quite surprised that one can find harder X-Sudokus than pure Sudokus. (Well, it also depends on how SER is extended).
But my question was restricted to pure Sudokus, for the very simple reason that there has been extensive search for them (much more than I'm aware of for X-Sudoku).

[coloin]

Perhaps I would react by saying it has already been done, see eg here.....
Leren

I think that post is edited as those SE 12.5 puzzles were rated in error.

Mikes post some months later has a true sudoku-X SE 12.0

But what does SE 12.0 mean I wonder

[dobrichev]

I would say "bravo".

For the top ratings, SE uses a strange mixture of rating components, that depend on involved classical methods tabular rating, the nesting complexity, chains' lengths logarithm. On top of this different rules apply where something is odd/even.
There is hardcoded limit of nesting depth, as well as for techniques used in nesting. This causes an average pencilmark-only sudoku to be beyond it's rating capabilities.
There is no rating upper limit.
Eventual harder puzzle can either reach the nesting limit and therefore be unsolvable for SE, or can be rated 12.0 or even more. Up to the natural limitations of maximal chain sizes which surely aren't accounted in the rating calculation algorithms.

[Pupp]

It's been mentioned multiple times that there are (at least one anyway), X-Sudoku puzzle(s) rated over 12 by SE.

I'm thinking jigsaw puzzles could get over 12 SE too. I think there are not a lot of puzzles outside of Sudoku, X-Sudoku, and perhaps Jigsaw puzzles that people really want to find the hardest possible puzzles for. And there are so many possible variation, even outside of the variations normally supported by SE and other Sudoku apps, that it's mostly pointless to try and find the hardest possible puzzles for most variants. Even inside of SE, I think few bother deep diving to find super difficult puzzled for variants other than the three I mentioned.

I"m only talking about standard sudoku puzzles and varients of those, and not variants that use math, like Killer Sudoku or Sum Sudoku, which are usually supported by different apps than normal Sudoku variants, or other Sudoku variants that don't fit in any particular genre.

[Pupp]

Here's an acronym: SCFLXPAD that isn't related to Sudoku.

Can anybody guess the acronym?

[Pupp]

Perhaps I would react by saying it has already been done, see eg here.....
Leren

I think that post is edited as those SE 12.5 puzzles were rated in error.

Mikes post some months later has a true sudoku-X SE 12.0

But what does SE 12.0 mean I wonder

Someone can post a link that explains how Sudoku Explorer (SE) rates puzzles. It's pretty comprehensive.
Essentially it's breaks down techniques by difficulty, so certain techniques would be a level 1 technique, some would be level 2, etc. I'm not sure, but I think the more you have to use techniques at the highest level difficulty in the puzzle will determine the 1/10th point. For example, if you had a puzzle with say, a Unique Rectangle, it might be rated 4.0 or something. (Just picking numbers out of the thin air for illustration purposes.) Perhaps the puzzle has a Unique Rectangle, an another technique rated in the 4.0 range, then it might rate the puzzle 4.1.

This is just for illustrative purposes, I have no idea what the algorithm does to get a specific rating for a puzzle, other than if it has at least one technique rated a certain level, then the puzzle has to be rated at least that much.

With Sudoku Explorer, I don't think it's possible for a standard, no variant, puzzle to be rated over 12, since the hardest techniques in it's library appears to only be rated 11. When it gets to variant puzzles, and in particular I'll mention X-Wing, apparently there are techniques a person can use that are rated 12, which could not be used to solve a standard Sudoku puzzle.

[mith]

That isn’t the case - I’ve seen (and I’m sure anyone who has spent time working with hard-by-SE puzzles has as well) steps rated 12.0 or higher, SE just don’t need them to resolve the puzzle. (The highest I remember was a 12.6 I think?).

There’s no particular structure reason a puzzle wouldn’t have 12+ techniques; it’s just a question of *needing* such a technique. What it would mean is that for every remaining wrong candidate in the puzzle, there is no technique rated less than 12 that can eliminate it.

I wouldn’t be terribly surprised to see a 12.0 eventually. I’d be more interested in how it is found - whether it’s just continued application of current search techniques, or someone comes up with something novel.

[coloin]

That isn’t the case - I’ve seen (and I’m sure anyone who has spent time working with hard-by-SE puzzles has as well) steps rated 12.0 or higher, SE just don’t need them to resolve the puzzle. (The highest I remember was a 12.6 I think?).

There’s no particular structure reason a puzzle wouldn’t have 12+ techniques; it’s just a question of *needing* such a technique. What it would mean is that for every remaining wrong candidate in the puzzle, there is no technique rated less than 12 that can eliminate it.

I wouldn’t be terribly surprised to see a 12.0 eventually. I’d be more interested in how it is found - whether it’s just continued application of current search techniques, or someone comes up with something novel.

Well ... I was really asking what difference in 11.9 move and a 12.0 move ....

however does it mean therefore that in a 11.9 puzzle at the hardest stage there are one or more eliminations rated at 11.9 and every other eliminations at that stage are > 11.9. ...?

I do appreciate that SE doesn’t take into account newer eliminating techniques ....

[mith]

My post was more directed at Pupp.

But yeah, in a 11.9 puzzle there could be eliminations rated harder than 11.9. (I have not actually stepped through any of the 11.9s to see if that's the case. I don't remember off the top of my head which puzzle I was looking at when I saw the 12.6 - one of mine, I assume. I do remember that SE froze on me when I tried to show it on the GUI.)

[mith]

So just to quickly verify that, I plugged in champagne dry since we know it's going to start with an 11.8.

SE (1.2.1 mode) gives 164 hints - two are 11.8, four are 11.9, and everything else is higher.

The rating is a base value + some measure of how complex the chain is. If I'm reading the code correctly, D*FC+DFC has a base of 10.5. So really what we're talking about is a complexity rating of 1.5+, vs. the max of 1.4 that has been seen.

For comparison sake, D*FC+MFC has a base of 10.0; I've found puzzles with a hardest technique in D*FC+MFC that rate anywhere from 10.7 to 11.7 (so a complexity rating of 0.7 to 1.7). So a 12.0 certainly isn't out of the question as far as the complexity of the chain... it's just a question of whether there a puzzle out there that can't be solved without using such a complex chain.

[mith]

SE finally unfroze after I clicked on the last hint, heh. 12.7, with a ridiculous 313 "views" (chains and nested chains). I pasted the explainer text into office, it's 331 pages.