denis_berthier wrote:.
BTW, the new 11.9 has "only" 209 candidates after Singles and whips[1].
Any search strategy for finding the highest SER that relies on additional pre-filters may fail to find some of the highest.
The recent findings showed this has been the case with adding some criterion related to the q1 or q2 ratings.
But it'd also be the case if one keeps only puzzles with too high numbers of candidates as starting points for proximity search.
Indeed, my previous q2 filters would have easily missed this one (49386 q2, 46752 q1).
(The correlation comment was just that the new 11.9 is 26c, expandable to 30c, vs. the 21c-23c previously known ones. The higher clue counts were just not well explored previously.)
champagne wrote:Great job. My last update of my own data base is now 2 years old.
From what I know of Sudoku Explainer, to produce such high ratings with more clues, you must have a highly interleaved system. This is usually linked to an exotic pattern.
Yes, the initial breakthrough on the high clue front was finding the trivalue oddagon pattern ("Thor's Hammer"), pointed out by marek previously in this thread. Essentially all of the high clue puzzles found have shown some variant of the pattern, and there have been a huge number of them from that initial seed.
In all cases I've looked at, there is a pattern of 6 digits in three boxes (after basics), one of which sees the other two, such that the empty cells span all three rows and columns, along with a fourth box (completing the rectangle of boxes) with 4 or 5 digits, such that some completions of the fourth box are impossible due to the parity of the permutations of missing digits. It shouldn't be all that hard to search for in existing puzzles, I just haven't written code to do it yet. It may also be possible to use the pattern as a starting point for generating puzzles, vs. neighborhood searching on existing puzzles.
Aside from the trivalue oddagon "exotic" pattern, there have been some other more specific non-3-colorable patterns discovered (shye's Patto Patto pattern being the one that comes to mind), so a generalized search for patterns with chromatic number 4 could be fruitful.
With so any fresh stuff, I intend to mark a pause in what I am doing an to scan your (an others) new findings.
Do you have a summary of all 11.x up to date somewhere?
I have been planning to get an update posted soon anyway. I still need to find some time to process jovi_al's two puzzle batches (a bunch of 19c puzzles, including some new 10.X, and a large collection of DJE puzzles as yet unchecked), as well as the pattern game. I think it may also be wise to let the neighborhood search iterate a bit longer with the new finds - I woke up this morning to a new batch of
139 11.8s (!), which more than doubles what was in the database. (I'll post those shortly.)
EDIT I checked my last data base of potential hardest I can add that no 19 had been found at that time with a rating >=11.0
and as far as I know, the main criteria for fresh seeds for a vicinity search has been when I worked on this topic the SER of old grids.
The high for 19c remains at 10.5 (three puzzles using gsf minlex, though I am aware of at least one coloin puzzle which is a 10.5 with the right morph). When I publish an update, I will include any 10.2+ from 19c and 35c-36c, since no 11s have been found there. (Edit while typing - I hadn't checked in a while, but I did finally generate some minimal 37c, two of which are skfr 10+. I'll have to rate those as well.)