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 denis_berthier » Wed Aug 30, 2023 12:02 pm

eleven wrote:Denis,
thanks for your stats, i had missed them. Since i don't have any cpu resources, i can't check it myself. Whenever i tried randomly a puzzle from a posted list, it was easy to solve, and i thought, those you had posted in the puzzles thread, were the hardest (all were manually solvable).

The idea of the Puzzles thread is to propose puzzles solvable by human players. Therefore, I selected carefully those I proposed. They are not the hardest, but the easiest (except a few ones).

eleven wrote:So "only" 40% of 3 mio (?) puzzles with extremely high ratings are easy to solve and, as champagne noted, a big part of of the rest can be (easily) essentially simplified to a lower rating.

I wouldn't say the 40% are easy. They require chains of length 5 that only few real players can find.
champagne's claims are his own imagination; they rest on no serious analysis.
denis_berthier
2010 Supporter
 
Posts: 4190
Joined: 19 June 2007
Location: Paris

Re: The hardest sudokus (new thread)

Postby champagne » Wed Aug 30, 2023 1:13 pm

denis_berthier wrote:champagne's claims are his own imagination; they rest on no serious analysis.


The only "claim" of champagne is to bet that none of these sudokus with the TH pattern can compete in the hardest sudokus hunt.
A counter example will be welcome :D :D
champagne
2017 Supporter
 
Posts: 7453
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby denis_berthier » Wed Aug 30, 2023 2:13 pm

champagne wrote:
denis_berthier wrote:champagne's claims are his own imagination; they rest on no serious analysis.

The only "claim" of champagne is to bet that none of these sudokus with the TH pattern can compete in the hardest sudokus hunt.
A counter example will be welcome :D :D

That's your claim or bet, not mine. Do your homework.
denis_berthier
2010 Supporter
 
Posts: 4190
Joined: 19 June 2007
Location: Paris

Re: The hardest sudokus (new thread)

Postby Paquita » Wed Aug 30, 2023 4:26 pm

Denis,

This is the new version of the non-T&E(3) puzzles.

Most of the puzzles are from mith or hendrik, some from me and JPF, and mith checked those for T&E(3). All in this file are not in miths collection of T&E(3). The only ones that have not been checked are my recent puzzles. So in theory it is possible that some of those are T&E(3), but they are not in miths collection either. It is unlikely that I did find some new T&E(3) puzzles, but not impossible. I will check those with your tool, that will take some time.

Yes they are all different. Mith and Hendrik contributed a lot.
I have now sorted them for the first rate. There are different formats for the rates and I did not fully sort them.

https://drive.google.com/file/d/1e2eVXOCkoU51NNbJtOY60DJk6khpO6Tl/view?usp=sharing
Paquita
 
Posts: 132
Joined: 11 November 2018

Re: The hardest sudokus (new thread)

Postby champagne » Wed Aug 30, 2023 4:46 pm

denis_berthier wrote:
champagne wrote:
denis_berthier wrote:champagne's claims are his own imagination; they rest on no serious analysis.

The only "claim" of champagne is to bet that none of these sudokus with the TH pattern can compete in the hardest sudokus hunt.
A counter example will be welcome :D :D

That's your claim or bet, not mine. Do your homework.

let people claiming to have the hardest sudoku do it :roll:
champagne
2017 Supporter
 
Posts: 7453
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby mith » Wed Aug 30, 2023 5:37 pm

champagne, no one here is claiming to have the hardest sudoku. Some are posting high SER puzzles, as that was for years and years the only real "standard" in this thread. The existing ph2010 database is ordered by SER, with no indication of bypasses by exotic techniques, so it's hard to fault anyone for that.

------

There are certainly puzzles in the te3 database which retain their SER after applications of TOFCs. This one is "only" an 11.0, just one I happened to have saved to look at more closely (specifically, it has guardians in all four boxes, exactly the cells of the 4-cycle in the TH pattern):

Code: Select all
........1..2..134...3.452.6...16.......2.3..523.........1....64.7.5.....89.4.....  ED=11.0/1.2/1.2; minimal of min-expand ID 89616


YZF manages three eliminations from TOFCs before eventually resorting to brute force. Maybe there are more eliminations that YZF isn't catching yet, I dunno.

But this is just one example. There's no particular reason to expect there won't be examples at higher SER (if there aren't already, I haven't checked for this specifically). And at some point searching for these involves a subjective choice of how difficult an OR-branched X-guardian TOFC with ALS nodes is compared to the various DFC levels, for various values of X. (Just as subjective choices would have to be made for Exocets, SET/MSLS/Multifish, etc.)

Anyway. To restate my main overall points here:

1. First and foremost, there is nothing to "repair" when it comes to the ph database and the TH puzzles. None of these puzzles exist in ph2010. The complaint here seems to be that people are posting new 11.9s and such in this thread without considering how easy they are to solve with TH; fine. I am posting my own findings in the other thread, and have really only been responding here to clarify what has already been found to avoid duplicating work.
2. The existing ph2010 database has already become a high SER database rather than a "true" hardest list (for whatever definition of hardest you prefer), and this happened long before the first TH puzzle was found. If the goal is to return to the purity of a list of puzzles that manual solvers having knowledge of exotic techniques would still struggle with, that part of the old hardest list (however small) should be removed first (as well as those with comparatively low ratings), and then newly found high SER puzzles can be evaluated; not all of those I've found and/or posted since ph2010 are TH based, I don't know about anyone else's. Those that are TH based are easily filtered and can be put to the side for further evaluation (to determine whether they are actually simplified or not by application of TH/TOFCs).
3. My own opinion is that the high SER database should be retained as a high SER database, just with additional fields to denote bypasses.
mith
 
Posts: 996
Joined: 14 July 2020

Re: The hardest sudokus (new thread)

Postby champagne » Wed Aug 30, 2023 6:00 pm

Hi Mith,

short an quick answer.

Yes the PH data base compiled puzzles based on the SER rating, the main tool available to detect "potential hardest puzzles"
Yes a great number of these puzzles have been identified as containing an "exotic pattern" with various effect on the SER after effect of the pattern. Most of them are flagged in my data base, but this is not up-to-date and they stay in the data base for various reasons (including the control to avoid duplicates)


I reached 81, time to lower my activity in this field, so I stopped the update of the data base and I am just a follower in the amazing discovery of the Th pattern.

I am quite sure that the search of variations in the TH pattern is of value. Less interested in the hunt for the highest SER rating (nothing to do with the fact that by chance, I still have the the highest one). The fact that we are here with a higher number of clues open the door for adjustments in the rules to avoid redundancy or quasi redundancy in the data base.

And I am surely interested to see the puzzles having a TH and still very hard either to reach the use of the rule of after use of the rule.
champagne
2017 Supporter
 
Posts: 7453
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby marek stefanik » Wed Aug 30, 2023 7:31 pm

Mith, it was exactly my point that the ways of spotting multifish and applying SET are very similar (with a minor difference in the order things are done in) with MSLS being quite similar as well. I wrote it in response to your statement that:
Viewing this as MSLS/Multifish involves looking at a lot of pencilmarked cells

I also don't think that anyone is faulting you (or anyone else), for TH puzzles taking over the thread, Eleven is just pointing out the problem.

Denis, it seems that you are (deliberately) ignoring the gap between what you think is solvable manually and the hardest puzzles found. If TH-based relations reduce a puzzle into T&E(1), it is nowhere near the hardest.

As for the above puzzle posted by Mith, here is the state after basics:
Code: Select all
,--------------------,--------------------,---------------------,
| 4679   468   46789 | 36789  23789 #2789 |#5789    5789   1    |
| 5679   568   2     | 6789  #789    1    | 3       4     #789  |
| 179    18    3     |#789    4      5    | 2      #789    6    |
:--------------------+--------------------+---------------------:
| 4579   458   45789 | 1      6     #4789 |#4789    23     23   |
| 14679  1468  46789 | 2     #789    3    | 46789  #789    5    |
| 2      3     46789 |#789    5      4789 | 146789  1789  #789  |
:--------------------+--------------------+---------------------:
| 35     2     1     | 3789   3789   789  | 5789    6      4    |
| 346    7     46    | 5      12389  2689 | 189     12389  2389 |
| 8      9     56    | 4      1237   267  | 157     12357  237  |
'--------------------'--------------------'---------------------'
Suppose 4r4c67.
The 4|6 in r8c3 is forced into r5c12 with 1, the other 4|6 is then forced into b4p369 => –46r19c3.
5r9c3, 5r7c7, 3r7c1, 13b8p58, 2r1c5
Now the 4 at the rectangle is the only guardian of the TH => 789 must each appear at R=r14c67.
789: c3R \ r14b4 => –789r4c12
Both cells are forced to be 5s, i.e. contra.
After that the puzzle solves easily with the remaining guardians.
I would be surprised if a puzzle that gets reduced to B3B S2B4B? (or lower) by TH-relations could be considered among the hardest, although I don't know how many puzzles are above that level and how many of them can be simplified.

Of course, sometimes it's difficult to draw the line at what relations should be used – take Kolk for example. Obviously we use the almost sk-loop and the almost PLQ, but my solution from last year (which I should really get ready to post, but I'm so unhappy with my path from 10.5 skfr onward), extensively uses that r7c89 cannot be 79. Would we want a computer solver to use it when estimating the puzzle's difficulty or is that too far?

Marek

Added: With 2r1c6=5r1c7 we get the following AIC:
5r7c1 = 5r7c7 – 5r1c7 = 2r1c6 – 2r1c5 = (12–3)r89c5 = 3b8p12 => -3r7c1
After basics, we see a 7|8|9 in r1c3 and b4p12, eliminated from the rectangle, so 2 and 5 are both needed (as 789 cannot repeat there).

I also checked with YZF_Sudoku and it used a more powerful step instead of the AIC: the 7|8|9 in r4c7 is forced into r7c456 and r789c6, so into r7c6. If there also were a 7|8|9 in r1c7, it would be forced into r7c6 in the same way, i.e. contra., hence -789r1c7

If instead of 4r4c67 we opt for eliminating 46r1456c3, we can do it in B2B with TH relations (including the OR relations between 25x at the TH rectangle for each x in 789).
Last edited by marek stefanik on Thu Aug 31, 2023 2:47 am, edited 2 times in total.
marek stefanik
 
Posts: 359
Joined: 05 May 2021

Re: The hardest sudokus (new thread)

Postby eleven » Wed Aug 30, 2023 10:53 pm

Nice move, but
marek stefanik wrote:After that the puzzle solves easily with the remaining guardians.

i can't see that. Please show it, too.
eleven
 
Posts: 3150
Joined: 10 February 2008

Re: The hardest sudokus (new thread)

Postby mith » Wed Aug 30, 2023 11:03 pm

marek stefanik wrote:I would be surprised if a puzzle that gets reduced to B3B (or lower) by TH-relations could be considered among the hardest, although I don't know how many puzzles are above B3B and how many of them can be simplified.


B3B would be pretty typical of an arbitrary SER 11.0, I think.

Incidentally, removing those guardian 4s does not simplify it according to SER (even the SukakuExplainer fork); SE just doesn't have all of the tools to make it "solve easily". (YZF can solve it in "orange", FWIW - meaning it doesn't need DFCs, basically.)
mith
 
Posts: 996
Joined: 14 July 2020

Re: The hardest sudokus (new thread)

Postby mith » Wed Aug 30, 2023 11:43 pm

Out of curiosity, I did a quick check of all the 11.9s against YZF.

I believe all the known TH-based 11.9s have TH-1 available, and reduce to SER 7.2, 7.8, or 8.3. There is no particular reason (that I am aware of) to expect all TH 11.9s would have this property, though it does seem the current 11.8+ in the te3 database have fewer guardians on average than the 11.7s.

The ph2010 11.9s:

eleven;Patience - MSLS 16 -> SER 9.0
OW;2015_08 - MSLS 16 -> SER 9.2
eleven;Second flush - Weak Exocet -> SER 10.6
dob;12_12_03 - JExocet -> SER 10.8

eleven-Imam_bayildi - JExocet -> SER 11.8
eleven-Kolk - JExocet -> SER 11.8
tarek-Golden_Nugget - JExocet -> SER 11.8
GP-kz0 - JExocet -> SER 11.8
GP-champagne dry - JExocet -> SER 11.8

Clearly some of these are still very hard by this standard (though again this is only taking into account techniques supported by YZF, and I'm not even sure I have the latest version; it's not using the almost SK-Loop in Kolk that Marek mentions, for example).
mith
 
Posts: 996
Joined: 14 July 2020

Re: The hardest sudokus (new thread)

Postby yzfwsf » Thu Aug 31, 2023 12:03 am

mith wrote:There are certainly puzzles in the te3 database which retain their SER after applications of TOFCs. This one is "only" an 11.0, just one I happened to have saved to look at more closely (specifically, it has guardians in all four boxes, exactly the cells of the 4-cycle in the TH pattern):

Code: Select all
........1..2..134...3.452.6...16.......2.3..523.........1....64.7.5.....89.4.....  ED=11.0/1.2/1.2; minimal of min-expand ID 89616


YZF manages three eliminations from TOFCs before eventually resorting to brute force. Maybe there are more eliminations that YZF isn't catching yet, I dunno.


If the Impossible Pattern Frocing Chain is added, this puzzle will become a puzzle that does not require brute force. In "Option" menu check "Search IPC".
yzfwsf
 
Posts: 900
Joined: 16 April 2019

Re: The hardest sudokus (new thread)

Postby mith » Thu Aug 31, 2023 12:21 am

Nice, I'll definitely try this out. (I was still on v624)
mith
 
Posts: 996
Joined: 14 July 2020

Re: The hardest sudokus (new thread)

Postby denis_berthier » Thu Aug 31, 2023 3:58 am

marek stefanik wrote:Denis, it seems that you are (deliberately) ignoring the gap between what you think is solvable manually and the hardest puzzles found. If TH-based relations reduce a puzzle into T&E(1), it is nowhere near the hardest.

As I don't know what "hardest" means and you probably don't know either, I can hardly answer.
I know what "deepest T&E-depth" means (that's what you find in mith's T&E(3) database), I know what SER-hardest means (that's what you find in the ph database).
All the rest is pointless discussions about undefined words. There's no "natural" or "human-based" rating.

If a puzzle is in T&E(3), it remains forever in T&E(3), even if it can be reduced by some tridagon rules to one in T&E(1).
If a puzzle has SER 11.9, it keeps forever its SER 11.9 rating, whatever rules you can use to simplify it.

My position is clear (and I think identical to mith's):
- there's a T&E(3) database (not by definition a tridagon database, even though all the puzzles in it happen to have non-degenerate tridagons);
- there's a SER-hardest database (not a "human hardest" database); this is and has always been the reality of the ph database.

Only with such clear definitions can one talk about the solvability of the puzzles in the databases.
If mith hadn't kept low SER puzzles in his T&E(3) database, no one could have found that after applying the tridagon-chain rules, the difficulty of the result doesn't depend on the original SER.

(as a secondary point: I say SER, but I think we should switch to PGXplainer).
.
denis_berthier
2010 Supporter
 
Posts: 4190
Joined: 19 June 2007
Location: Paris

Re: The hardest sudokus (new thread)

Postby denis_berthier » Thu Aug 31, 2023 4:01 am

mith wrote:B3B would be pretty typical of an arbitrary SER 11.0, I think.

Yes.
Table 11.2 in [PBCS] gives the vague correspondance between SER and BpB. SER 11.3 implies 70% "probability" in B3B (computed on a small and biased sample, so this is just an indication)
.
denis_berthier
2010 Supporter
 
Posts: 4190
Joined: 19 June 2007
Location: Paris

PreviousNext

Return to General