The hardest sudokus (new thread)

Everything about Sudoku that doesn't fit in one of the other sections

Re: The hardest sudokus (new thread)

Postby champagne » Mon Aug 28, 2023 6:39 am

denis_berthier wrote:... which doesn't mean the T&E(3) domain is inactive:.

right, but mainly here and occupying more and more space!!

denis_berthier wrote:I think huge threads like the "hardest" one lead to undefined goals. What you call hijacking of them is only the result of these undefined goals.


Creativity requires flexibility, but hijackers push the ball so far that the original goal can be hidden. This does not say that the original goal was not clear;
champagne
2017 Supporter
 
Posts: 7453
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby denis_berthier » Mon Aug 28, 2023 6:47 am

champagne wrote:
denis_berthier wrote:... which doesn't mean the T&E(3) domain is inactive:.

right, but mainly here and occupying more and more space!!

If you mean anything about T&E(3) puzzles should be in the T&E(3) thread instead of here, I agree.

champagne wrote:
denis_berthier wrote:I think huge threads like the "hardest" one lead to undefined goals. What you call hijacking of them is only the result of these undefined goals.

Creativity requires flexibility, but hijackers push the ball so far that the original goal can be hidden. This does not say that the original goal was not clear;

Well, then why is it impossible to find such a goal written anywhere in this thread?
denis_berthier
2010 Supporter
 
Posts: 4190
Joined: 19 June 2007
Location: Paris

Re: The hardest sudokus (new thread)

Postby champagne » Mon Aug 28, 2023 7:10 am

denis_berthier wrote:Well, then why is it impossible to find such a goal written anywhere in this thread?


The first thread (lost) has been open long long ago and I was not in the forum at that time, but the context was very clear. May-be tarek and JPF could comment on that.
At that time, One claimed to have produced the "hardest suoku". All the discussions were on manual solving.
champagne
2017 Supporter
 
Posts: 7453
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby denis_berthier » Mon Aug 28, 2023 8:09 am

.
For anyone interested in the real origins of the "hardest sudokus" thread, here's Ravel original post (dated 28 May 2006):
http://forum.enjoysudoku.com/the-hardest-sudokus-t4212.html?hilit=hardest%20sudokus
Skip the edits. Then you'll see it's all about ratings by programs (including SE) that don't have much in common with manual solver's techniques.
The last post in this thread (by Tarek) is dated 09 Jan 2009, 04:59. It says:
"I will be starting a new thread -- The hardest sudokus (new thread) -- with the 1st post having the list of puzzles in a similar fashion to the one currently on wikipedia.
This is mainly due to the the fact that ravel is no longer an active member of this forum, with so much going on since his departure."
(There are a few post after that one, but they've been added later.)

The first post (by Tarek) in the new thread is dated 09 Jan 2009, 05:03, in perfect continuity with the last one of the old thread.
What's certain is, the first post of the new thread is really the 1st post in it, nothing has been lost in this thread: the 2nd post says explicitly it is the 2nd post.
And what can we see in the first post? It considers a list of "popular ratings", none of which is related to manual solving.
What's in the 2nd? A list of "Popular downloadable rating programs".


I don't care much about who is right or wrong. What I insist on is, without clear goals, don't expect to have a thread that doesn't go in all directions. The quickly developing T&E(3) domain brings up the necessity to state the goals of this "hardest" thread. It may be the opportunity to close this overly long thread and to open a new one with precise goals stated at the start. As mith appears to be the one who will manage this database and do the hard work, it's his prerogative to decide.
.
denis_berthier
2010 Supporter
 
Posts: 4190
Joined: 19 June 2007
Location: Paris

Re: The hardest sudokus (new thread)

Postby eleven » Mon Aug 28, 2023 7:25 pm

Fact is, that Ravel was only interested in hardest puzzles concerning manual solving. He tested various ratings and let the manual solvers try the found puzzles. His conclusion was, that (at that time) SER fitted best for his purpose. Since then it became standard for testing extremely hard puzzles.
Now we know, that a lot of easily solvable puzzles have been discovered, which - all due to the same pattern, which can be easily spotted - have extremely high SER and T&E ratings, so those don't fit anymore for Ravel's original purpose - and we have no program, which does it. That's really a pity.
Since my post is now hidden again by long answers, i want to repeat: Puzzles, which can be solved on paper in 15 minutes are displaced in a "hardest sudokus" thread.
And a call to all interested sudoku programmers: Please try to write a better rating program, from scratch or by improving existing ones.
eleven
 
Posts: 3150
Joined: 10 February 2008

Re: The hardest sudokus (new thread)

Postby mith » Mon Aug 28, 2023 10:20 pm

This issue has existed for a long time, well before the TH pattern and depth 3 puzzles were discovered. Just as with MSLS/SET and Exocets, the trouble is that it's impossible to define a single rating for puzzles with vs without these techniques without making some arbitrary choice about their difficulty. The SER +0.5 levels for chains are arbitrary as it is.

The TH/trivalue oddagon issue is exacerbated by the fact that just having the pattern of cells does not guarantee the difficulty of the deduction from that pattern. There are definitely puzzles which rate SER 11.8+ that are solvable in a few minutes by a human solver familiar with the technique. There are also some puzzles that only give the potential for a large OR-branching chain (or equivalent). Never mind defining what it means to "have a TH pattern" being difficult to begin with - most (all?) puzzles will have some TH pattern with a huge number of guardians, never mind other chromatic patterns.

I do at least have a script for finding the "special" TH cases (at most 1 of three digits as givens, all confined to a single box, with the diagonal pattern in the four boxes not seeing that box), and I could mark those in the database (obviously pointless for the depth 3 database until we find one that doesn't have it, not pointless for the ph database). I also have a way to identify puzzles solvable with only rank 0 deductions (including MSLS/SET), but it's too slow to run on a large collection.

The point is: SER + TH isn't an answer. SER + TH + SET + Exocets + ALS techniques + Finned Fish + Fireworks + ... is better in the sense of being more complete with current human solving but would be a disaster when it comes to arbitrary rating decisions. (YZF already incorporates all of these and more; so maybe the metric for the ph database should be "does YZF require brute force". It only goes up to level 0 DFCs though, so that would be a very inclusive database in terms of SER.)
mith
 
Posts: 996
Joined: 14 July 2020

Re: The hardest sudokus (new thread)

Postby eleven » Tue Aug 29, 2023 12:14 am

SET, Exocet and fireworks are not easy to spot, different to TH with single extra candidates, which make so many high rated puzzles trivial. So in a first step all the latter have to be swiped off from each hardest list, followed by those, where 2 extra candidates give a link, which can easily be used to solve it etc.
How high detecting such patterns is rated, may be a matter of taste, but as it is, it's just obviously ridiculous.
eleven
 
Posts: 3150
Joined: 10 February 2008

Re: The hardest sudokus (new thread)

Postby mith » Tue Aug 29, 2023 3:08 am

eleven wrote:SET, Exocet and fireworks are not easy to spot, different to TH with single extra candidates, which make so many high rated puzzles trivial. So in a first step all the latter have to be swiped off from each hardest list, followed by those, where 2 extra candidates give a link, which can easily be used to solve it etc.
How high detecting such patterns is rated, may be a matter of taste, but as it is, it's just obviously ridiculous.


It's *very* easy to spot a row/column SET in most cases; and SET can definitely make high rated puzzles trivial. Exocet can be viewed as a rank >0 extension of SET. There are other patterns with varying degrees of difficulty for a human solver, but if the goal of a "potential hardest" database is to only have puzzles that are hard for a human to solve, a puzzle solvable by a single fireworks is nowhere near that level.

I absolutely agree that puzzles with TH-1 exist that make high SER puzzles trivial. There are more of these known because I have been searching this space. I have published with my last update (and will with the coming one) classifications of the depth 3 puzzles by guardian count. There are also TH-1 puzzles where the application of TH-1 does not make the puzzle trivial. There are puzzles with TH-2 that are relatively easy to get deductions from, and those that aren't. There are all sorts of TH puzzles that are easier to solve with relabeling (I don't know that anyone has checked the ph database for non-TH puzzles that get easier with relabeling, but there are certainly some). There are some that just aren't easy.

Any attempt to ignore puzzles that are high SER requires a line to be drawn. "Obviously" that line should at least exclude puzzles which have TH-1 and solve trivially afterward. Fine. But let's not pretend that the existing ph database consists only of hard-by-current-human-understanding puzzles. That hasn't been true for a long time. I am certainly open to changing the presentation of a future ph-style database to account for that, if I am the one who ends up maintaining it, but I don't know that there will ever be consensus on what qualifies and what doesn't, and knowledge of solving sudoku continues to advance in the meantime.
mith
 
Posts: 996
Joined: 14 July 2020

Re: The hardest sudokus (new thread)

Postby Paquita » Tue Aug 29, 2023 8:57 am

Mith,

This is my file with non-T&E(3) puzzles. It includes the ph2010 (11.6plus) and what I scraped since. Puzzles are named, maxlex, and rated. Maybe it is useful for you when you get to the follow-up of ph2010 for non-T&E(3)

https://drive.google.com/file/d/1QYOfqBfOBQbD7VPaSPvR3AgntDKrRbTe/view?usp=sharing
Paquita
 
Posts: 132
Joined: 11 November 2018

Re: The hardest sudokus (new thread)

Postby Paquita » Tue Aug 29, 2023 9:19 am

Oeps..these are not in my file

98.76.5..7.59.4....46.859..56.......4.78.6.5..98.........6.7......4...32.......9.;11.9/2.0/2.0 monh
98.76.54.7.5.......6.8..7..8.....6..5.6....97.47...85.4.9..........394......2....;11.9/1.2/1.2 monh
98.76.5..7.5..98...64..5....96.......4.....32.....4.8..79.46.5....95.......8.7...;11.9/1.2/1.2 monh
98.76.5..7.59.4.....6.859..56.......4.78.6.5..98........4....9....6.7......4...32;11.9/1.2/1.2 monh
98.76.54.7.5.......6.9..7..89....6..5.6....87.47...95.4.8..........384......2....;11.9/1.2/1.2 monh
98.76.54.7.5.......6.9..7..89....6..5......87.47..695.4.8..........384......2....;11.9/1.2/1.2 monh
Paquita
 
Posts: 132
Joined: 11 November 2018

Re: The hardest sudokus (new thread)

Postby Paquita » Tue Aug 29, 2023 10:06 am

Oh its no mistake...those 11.9's are in T&E(3)
Paquita
 
Posts: 132
Joined: 11 November 2018

Re: The hardest sudokus (new thread)

Postby eleven » Tue Aug 29, 2023 10:19 am

Mith,
you are of course right, that there will never be a perfect rating system. Thank god, this fact is most important to keep the attractiveness of this thread. All the time the best solvers tried to crack high rated puzzles with new methods, and became famous in the community, if they had success with the one or other puzzle.
Champagne had an own thread with "exotic patterns", which allowed to solve or simplify many (by far not a majority) of known hard rated puzzles. Only experts could apply them without the big help of programs.

But things changed with the one pattern you have found. Any newspaper puzzle solver can learn it in 15 minutes, and suddenly is able to solve thousands of highest rated puzzles.
On the other hand the search for new etxremely hard puzzles only concentrated on this pattern, simply because it was the easiest way to find a SER 11.8+ or a T&E(3), regardless, if it was trivial or hard to solve.
No doubt, that you had made a great discovery, and we saw, that many TH puzzles have been a big enrichment for great manual solvers too.

But in my eyes we have to find a way to come back to a hardest database, which earns it's name, be it by a new rating or by throwing out all the puzzles, which are essentially simplified by easy-to-spot TH-moves ([edit - mistook a puzzle] e.g. all of Paquita's 11.9's above).
Last edited by eleven on Tue Aug 29, 2023 12:39 pm, edited 2 times in total.
eleven
 
Posts: 3150
Joined: 10 February 2008

Re: The hardest sudokus (new thread)

Postby Paquita » Tue Aug 29, 2023 11:12 am

Oh and in my file are also these but they are non-minimal

Code: Select all
98765....64....98..3298....8..59.14.4...186.9...4.6.58.148..59...8.4.......1..8.4; ED=11.7/11.7/3.4 mith
98765....64.73....2.3...76.12.36.5.7.6..2.......1..62...657.1.2..2.1637....2.3.56; ED=11.7/11.7/3.4 mith
98765....6438.....251...86.8....9...1..5.86.2...1..98..6.98.12..182.5.96....165.8; ED=11.6/11.6/3.4 mith
98765....6438.....251...86.8....9...1..5.86.2...1..98..6.28.19..189.5.26....165.8; ED=11.6/11.6/3.4 mith
98765....64.39.....32...96.3...6..522....53.6....2319..5321.6.9.295.6.......3952.; ED=11.7/11.7/8.3 mith
98765....64.93.....32...96.3..2.5.912....65.3...3..62..531.92.6.2956........23.59; ED=11.7/11.7/8.3 mith
98765....64.93.....32...96.3..2.56.12....9.53...3..29..531.6.29.2659........235.6; ED=11.7/11.7/8.3 mith
98.76.54.7.58.4....46.95...6.7.....9....7.3.........2...954...8...6.7.......89..4; ED=11.8/11.8/3.4 non-minimal mith
98.76.54.7.58.4....46.95...6.7.....9....7.3.........2...958...4...6.7.......49..8; ED=11.8/11.8/3.4 non-minimal mith
9876........59.........4...86....3..7.2.......398.627.6.8.2..97..3.6.8.2......63.; ED=11.8/2.0/2.0 mith
9876........59.8.......4...83.......7.6...2...927.638.67..3..98.2..6.7.3......62.; ED=11.8/2.0/2.0 mith
987.........96.8.......5...84.......7.3...2...92.7348.37.4...98.2.3..7.4......32.; ED=11.8/2.0/2.0 mith
9876........59.........4...86....3..7.2.......398.627.6.8.2..972.....63...3.6.8.2; ED=11.8/10.5/2.6 mith
9876........59.8.......4...83.......7.6...2...927.638.67..3..983.....62..2..6.7.3; ED=11.8/11.8/2.6 mith
98.76.5..7.5.48....465.9....974.6.8....95........87..9.6......3..4...6.2...6.....; ED=11.8/2.0/2.0 mith
98.76.54.75..84......9.58.737....4..2......5...6........98.6.7....54.98.......6..; ED=11.8/11.8/2.6 mith
9876.........958.......4...8.32..69.29.3.67...76......7.9...26.32..6..87.68...3..; ED=11.8/2.0/2.0 mith
9876.........958.......4...86....3..7.9...26..23.6..8767.......3.82..69..923.67..; ED=11.8/2.0/2.0 mith
9876.........958.......4...83.2..69.2.93.67...76......79....26.3.2.6..87.68...3..; ED=11.8/2.0/2.0 mith
9876.........958.......4...8.32..69.29.3.67.8.76......7.9...26.32..6..87.68...3..; ED=11.8/11.8/2.6 mith
9876.........958.......4...86....3..7.9...26..23.6..8767.......3.82..69..923.67.8; ED=11.8/11.8/2.6 mith
9876.........958.......4...83.2..69.2.93.67.8.76......79....26.3.2.6..87.68...3..; ED=11.8/11.8/2.6 mith
98765.......4..........3...8.62..75.75.8.6..2.29....866.8.2..972...6.8.5..5......; ED=11.8/11.8/2.6 mith
98765.......4..........3...8.62..95.72.....86.598.6..26.8.2..795..........2.6.8.5; ED=11.8/11.8/2.6 mith
98765.......4..........3...8.62..75.75.8.6..2.29....866...2.8.52.8.6..97..5......; ED=11.8/11.8/2.6 mith
987.........65.............8.493.2..27........932...8472...38.93....247..4.....23; ED=11.8/11.8/2.6 mith
987.........65.............8.493.2..72........932...843....247.27...38.9.4.....23; ED=11.8/11.8/2.6 mith
987.........65.............8.493.2..72........932...8437...28.92....347..4.....23; ED=11.8/11.8/2.6 mith
987.........65.............8.493.2..72........932...8437...28.9.4.....23.....347.; ED=11.8/2.0/2.0 mith

Paquita
 
Posts: 132
Joined: 11 November 2018

Re: The hardest sudokus (new thread)

Postby Paquita » Tue Aug 29, 2023 12:55 pm

Oh and these must be contributed to JPF

987......6..54.............37.82.6..2.86...73..9.......9...27.8.2...639...3.....2;ED=11.8/1.2/1.2;JPF
987......6..54.............37.82.6..2.86...73.9........6...239...3.....2.....67.8;ED=11.8/1.2/1.2;JPF
987......6..54.............37.82.6..2.86...73.9.........3....62.....67.8.....239.;ED=11.8/1.2/1.2;JPF
987......6..54.............37.82.6..2.86...73.9........2...639...3....62.....27.8;ED=11.8/1.2/1.2;JPF
987......6..54.............37.82.6..2.86...73.9........6...27.8.2...639...3.....2;ED=11.8/1.2/1.2;JPF
98.76.5..7.5..9.6...4......6..98....5..4.7....78.56..4.9..........8...39.......2.;ED=11.8/1.2/1.2;JPF
98.76.5..7.5..8....64..9...4...87.9...76.4..8...95......6...43....4............2.;ED=11.8/1.2/1.2;JPF

I can give an updated file, if it is any use?
Paquita
 
Posts: 132
Joined: 11 November 2018

Re: The hardest sudokus (new thread)

Postby denis_berthier » Tue Aug 29, 2023 2:20 pm

Paquita wrote:I can give an updated file, if it is any use?

I you provide a file, it's of no use if it's not up-to-date.
denis_berthier
2010 Supporter
 
Posts: 4190
Joined: 19 June 2007
Location: Paris

PreviousNext

Return to General