Game over
We have 49158 ED sudokus, not one more.This has been a very long task done with a limited power.
The scan to extract all 17s having one band/stack with more than 6 clues started in the end of the summer 2017 and was finished in October 2018.
The scan for the distribution 566 656 (no stack over 6 clues) started after, with ” more or less” 20 cores of an I7 (processors available at this moment).
Mathimagics came in June 2020 to double the throughput and processed bands index 90 to 119 and 130 to 199. This was a difficult area and it has been a key contribution to the project.
Mith arrived later and took over the slice 120-129. Thanks to him as well.
Most of the scan for this distribution has been done with the V6 version of the code.
The proof that no 16 clues exists was done in one year with 820 cores. Here a rough estimate would give: the proof that all 17s are known was done in 5 years with less than 40 active cores. This is an amazing improvement in performance, due to blue’s ideas.
In 2019, Blue had written a 18/19 search much more efficient that the current scan. He sent to mathimagics a DLL analyzing a solution grid to extract possible 18s.
Unhappily, we don't have the sources of the DLL. Blue’s last post was end of 2019.
I worked on this topic, trying to crack blue’s fresh ideas. I have in hands a code approaching blue’s performance, still in debugging status.
If all this works, we have a sound basis to write a V7 much faster tan the V6 (too late) but also to think of a 18 scan producing all ED 18.
Next month, I’ll pass 80, and I have other heavy constraints. I had to reduce drastically the time dedicated to the sudoku field. The work on 18 clues will likely be my last search in this field.
The draft of the current code is here
sk_is18 It will be updated on a regular basis.
Then (in one month ??) I intend to prepare a V7 on the 17 search. Although the basic concepts are the same, the process organization is quite different. In the DLL, only one solution grid is studied (with six permutations in my code), in the V6 search, several bands 3 are attached to a pair {band1;band2}, and we have three versions of the code differing by the set of bands 3 attached and the distribution checked;
The reason to prepare this V7 is to have a file of known for tests and validation of the code. The final code that I would produce is the 18 scan, not far from the 17 scan.
Aging and health problems can stop this at any moment, so I’ll prepare comments on the 18 current program as soon as possible and I’ll keep the draft available in the appropriate repositories.
I’ll open a separate thread for the 18 clues search.
