The New Sudoku Players' Forum

[champagne]

Having thoughts to all of us locked at home, with may be some free time to read this, I open a thread dedicated to the rank 0 logic in hard puzzles.

As explained in the attached document, to avoid endless discussions with different contents of the “rank 0 logic” concept, I introduce the term TLG0 (Truth Links Group Rank 0) with precise constraints, but the TLG0 has been long discussed under the term “rank 0 logic”.

This reflects my current work on this topic. I introduce new rules for eliminations and some fresh ideas to build such TLG0
In parallel, I am revising my code using the tools described in the document.

I should be in a position, in April, to run the code on big files.

The attached document has been prepared with an old version of Microsoft Publisher. This is to-day my best tool.

The pages are loaded on my web site. This is a living draft. I’ll add post in due time when a significant progress has been made.

the entry in the document is here

I lock several post to have room for
The summary of the results of the search in big files
Comments on the code
...

[champagne]

to come

[champagne]

to come

[champagne]

to come

[champagne]

to come

[champagne]

This thread was open more than one week ago.
Meantime, I made changes in the process to have flexibility and statistics available. I faced a bug in the stack difficult to understand for me and lost some more time on another topic to be seen later in the exotic pattern thread.

I run the code on the first (highest ratings) 200k puzzles of the hardest data base with some limitations in the search especially for the MSLS TLG0 not yet fully transferred in the new frame.

all ratings are >=11.1 in this sample. I got the following results

Code: Select all

142153   something seen   

130811   JExocet   
124591   exocet not JE   

=====================

4340   TLG0 R   
1766   TLG0 C

=====================

2571   TLG0 2r2c   
5   TLG0 4B   
1735  2r2c node occupied   
22   TLG0 + empty rect in box   

820  Vloop   

=====================

3366   MSLS   

More than 70% have an exotic pattern recognized by the code. The ratio was much better for the first 100K puzzles.
By far, the main exotic pattern is the exocet (off topic in this thread). In the chunk 100K-200K, the ratio of puzzles having an exocet was unusually low.

Compared to the exocet pattern, recognized TLG0 are not so common.

In the first group, TLG0 R and TLG0 C, the search was limited to 4 rows/columns. The search was cancelled at the first seen.

The second group is of the family of the SKloop/Vloop.

Only 820 puzzles in the sample had a SKloop/Vloop. This is something already seen earlier. The SK loop has been long discussed, but it is not a very common pattern.

The MSLS search was limited to a 4x4 matrix.

In the conditions of this run, about 100k puzzles are searched in one hour on my laptop

[yzfwsf]

Is it to solve the puzzle step by step to confirm whether there is MSLS or to search for MSLS once after performing simple steps.

[champagne]

Is it to solve the puzzle step by step to confirm whether there is MSLS or to search for MSLS once after performing simple steps.

The basic idea is to activate the TLG0 search in due time. This means IMO after basic move at the minimum (till XYZwing in SE ratings).
In this run, I also apply in blind mode single floor (one digit) eliminations for 3 reasons

a) I am currently working on the potenital hardest file,
b) In the multi floors analysis, a "single floor elimination" gives a positive signal for every multi-floors containing this digit.
c) except in very special cases, a "single floor elimination" is rated below 8.2 in SE.

Your "small example" shows that you are working also in areas of lower difficulty.

And yes, the idea is to apply eliminations in the solver

[yzfwsf]

3trues 15b2,6n5 \ 3links 15c5,3n6

rank0.png (18.86 KiB) Viewed 1574 times

[champagne]

well done

the corresponding XY loop

1r6c5 - 1r3c5 = 1r3c6 -5r3c6 = 5r13c5 - 5r6c5 = 1r6c5 loop

and again an empty rectangle pattern for the digit 5

[Mauriès Robert]

Hi all,
My way of proceeding to these eliminations is the following (see the notion of anti-track in TDP):
-5r1c5->(4789)b2p2456->[(15)r3c56 and (15)r36c5] => -5r4c5, -789r3c6.
This exploits the almost quad (4789/5)b2p2456.
Robert

[champagne]

Hi all,
My way of proceeding to these eliminations is the following (see the notion of anti-track in TDP):
-5r1c5->(4789)b2p2456->[(15)r3c56 and (15)r36c5] => -5r4c5, -789r3c6.
This exploits the almost quad (4789/5)b2p2456.
Robert

Elimination of 5r47c5 is easy to understand. the almost hidden pair 15 in box 2 works as well.
Elimination of 789r3c6 is harder to justify (requiring some work on the digit '1'). It would be interesting to see how you exptess it within the TDP logic.

The logis expressed by" yzfwsy" is much simpler and the TLG0 much shorter than the equivalent XY loop.

to " yzfwsf":
is it a standard tool within your solver of a separate approach?
and this pushes me to think of a similar search.

[yzfwsf]

to " yzfwsf":
is it a standard tool within your solver of a separate approach?
and this pushes me to think of a similar search.

Yes, there's a technique in my solver called almost pair/triple.

[Mauriès Robert]

Hi Champagne.

Elimination of 789r3c6 is harder to justify (requiring some work on the digit '1'). It would be interesting to see how you exptess it within the TDP logic.

The direct elimination of 789r3c6, is done with a track generated by this set E=789r3c6 whose construction is based on the triplet (789)b2p469 :
P(E): 789r3c6->[(789)b2p469->5r1c5]->1r3c5 => empty r6c5 => invalid P(E) => -E.
Sincerely
Robert

[champagne]

The direct elimination of 789r3c6, is done with a track generated by this set E=789r3c6 whose construction is based on the triplet (789)b2p469 :
P(E): 789r3c6->[(789)b2p469->5r1c5]->1r3c5 => empty r6c5 => invalid P(E) => -E.
Sincerely
Robert

Hi Robert,
Here no problem (although, again, I prefer the hidden pair 15r13c5 in the logic). I thought that this was an elimination of the anti track.