The New Sudoku Players' Forum

[999_Springs]

Thread title says it all, pretty much. I was gone from here for like 3 years and now I find out that developments in exotic patterns (sk loop, (j)exocet, multi fish, symmetry of givens, stuff like that) mean that human players can often make progress in SE 10+ territory using these patterns. So, I would like to begin to learn these things. I'm looking through champagne's "resume of exotic patterns" thread slowly; it's pretty dense stuff to read, and any new posts on the topic are intractable to me. I'm looking for additional resources/links/stuff/explanations that can help beginners that I can read up on in my own time, preferably with an emphasis on manual solving, anyone help?

also a few questions.
- What's the usual success rate for these patterns? like, what % of SE 10+ can be solvable by some human player who knows how to do these things without resorting to AIC-nets with nested solving techniques?
- Is there any way to quickly determine when a puzzle can be cracked with methods like this, whether by actually trying to solve it, or even is there a freely available computer solver that can do these things? If not, can I get a status report on these 3 puzzles: top1465#77, AI escargot, AI's other puzzle (june 2012) (by status report, I don't mean complete solution path, just hints)
- what happens in SE 9.5ish territory? are we still confined to AIC nets or can we still use these methods? which would be simpler?
- when you're looking for an exocet how do you determine which 2 base cells with 3 or 4 candidates to start looking from and how do you search for a target? there are bound to be zillions of suitable starting cell pairs

any help would be appreciated. sorry if this sounds vague, i wrote this in a rush

edit: 350th post yay

[champagne]

If not, can I get a status report on these 3 puzzles: top1465#77, AI escargot, AI's other puzzle (june 2012) (by status report, I don't mean complete solution path, just hints)

I have many comments on your post, requiring more time.

Just to save time, could you post directly these puzzles here.

As far as I remember,

AI escargot
1) does not have a very high SE rating
2) had none of the exotic pattern identified when I checked it

[Pat]

• AI escargot
• AI's other puzzle (june 2012)

Just to save time, could you post directly these puzzles here.

Arto Inkala claims copyright,
so i just give links

• AI Escargot
  was posted on the forum
  by ArtoI (2006.Nov.3)
  
  
• another one (2012.Jun.27)
  posted as image

[ronk]

How about relocating this thread to the "Help with ... solving techniques" sub-forum?

Good thought, moved. JasonLion-Admin

[champagne]

I have been busy these days, so I start late to try to answer to your questions.

can I get a status report on these 3 puzzles: top1465#77, AI escargot, AI's other puzzle (june 2012) (by status report, I don't mean complete solution path, just hints)

Let's start with that. I loaded four puzzles corresponding I hope to your expectations

Code: Select all

7.....4...2..7..8...3..8..9...5..3...6..2..9...1..7..6...3..9...3..4..6...9..1..5;1465#77
1....7.9..3..2...8..96..5....53..9...1..8...26....4...3......1..4......7..7...3..;AI Escargot
8..........36......7..9.2...5...7.......457.....1...3...1....68..85...1..9....4..;inkala2
..53.....8......2..7..1.5..4....53...1..7...6..32...8..6.5....9..4....3......97..;inkala3


skfr ratings are respectively 9.6/1.2/1.2 10.4/1.2/1.2 10.6/10.6/9.6 10.5/1.2/1.2

With such ratings, only inkala2 meet the entry barrier to the file of potential hardest.

I looked for exotic patterns and found only one in inkala 2 in that position

Code: Select all

8      1246  24569 |2347  12357  1234  |13569 4579  1345679
12459  124   3     |6     12578  1248  |1589  45789 14579   
1456   7     456   |348   9      1348  |2     458   13456   
-----------------------------------------------------------
123469 5     2469  |2389  2368   7     |1689  2489  12469   
12369  12368 269   |2389  4      5     |7     289   1269   
24679  2468  24679 |1     268    2689  |5689  3     24569   
-----------------------------------------------------------
23457  234   1     |23479 237    2349  |359   6     8       
23467  2346  8     |5     2367   23469 |39    1     2379   
23567  9     2567  |2378  123678 12368 |4     257   2357   
SLG rank 0
16 Truths = {1C2567 3C2567 6C2567 8C2567 }
16 Links = {1r12 3r178 6r168 8r26 3n6 4n57 5n2 9n56 }
23 eliminations
 4r3c6 2r4c5 9r4c7 2r5c2 2r9c5 7r9c5 2r9c6 1r1c9 3r1c4 3r1c9 6r1c3 6r1c9 1r2c1
 1r2c9 8r2c8 6r6c1 6r6c3 6r6c9 3r7c1 3r7c4 3r8c1 3r8c9 6r8c1


at that point, you have a hidden pair 36 in box 3 and can continue with r2c7=1 r3c8=8
The highest rating from that point in serate is 9.0

- Is there any way to quickly determine when a puzzle can be cracked with methods like this, whether by actually trying to solve it, or even is there a freely available computer solver that can do these things?

Several other members of the forum have a program doing all or part of that, but i don't know what is the availability of their tools.

My code is public skmpp project
and reuses most of the skfr project code

but this is the mirror of my development platform so

I don't produce .exe files as in the skfr project (skfr .exe file has been downloaded about 60 times )
It has all qualities and defaults of such a tool

It is up-to-date unhappily including the last bugs introduced in the code (usually, I update the code when compilation is ok)
There is no file explaining the options of the command line. (nobody so far showed interest to justify it)

In that program, I have separate processes to find each of the exotic patterns.


But you also have experts in the manual search of such patterns.

The SK loop is usually easy to locate, but not very common
The JExocet has been well described by David P Bird and is not too hard to detect
Manual players are also efficient in the search of rank 0 logic or equivalent, but I can't comment for them.

[champagne]

- What's the usual success rate for these patterns? like, what % of SE 10+ can be solvable by some human player who knows how to do these things without resorting to AIC-nets with nested solving techniques?

This is a good but very open question.

The easiest is to start form facts.

The sudoku community has built over time a significant collection of "hardest puzzles". This has been done case by case during years till "eleven" made a systematic search in the 20/23 area and quickly increased the volume of such puzzles. Later on, I concentrated on that area and "dobrichev" came with a short but very productive investigation.

all these results are collected in a data base whose last public version is in my google drive

I continue to search for such puzzles and recently started the search in the 26 clues area. My current file has about 1 million puzzles, but my last version fully covered toward your question had 650k puzzles.

In that file (rounded figures)

Code: Select all

80% of the puzzles had an exocet pattern, mainly a JExocet
25% of these puzzles had a double exocet pattern
2.5 % had a SK loop
0.01% had a symmetry of given, a very confidential pattern
2.4% had a rank0 logic and none of the former patterns

So 85% of the file had a pattern easy to use as solving technique.

13.5% had a partial exocet, a field still widely unexplored.

This means that for about 3% of the file (1% is 6000 puzzles), we had no clue to go faster than through the rating process. The only new idea is to study complementary AAHS when they show up.

All these puzzles have been selected using serate rating process and rate at the minimum 10.3 diamond, 10.5 pearl or any ED>=11.0

The serate process is a chain oriented process. It includes a limited number of global objects as XWings, Pairs, but remains heavily chain oriented.

My conjecture is that a small number of global objects resisting to chains is responsible of the high ratings. The SK loop was the first found, the Exocet pattern came later and The rank 0 logic (where many things have not yet been found IMO) came later.

One consequence is that such a pattern pushes immediately the rating at a very high level and the chances to find it with lower ratings are relatively small. This is especially true for the Exotic pattern. The rank 0 logic offers many possibilities with a smaller number of sets, but most of the corresponding patterns already received a name (eg: sue de coq).


The other question is what is the situation left after the direct effect of the pattern has been applied. There is no general answer to that question and, having nearly 1 million of puzzles so far to analyse, it can not be answered without an appropriate program.

My own program is not yet ready to do such a work. I hope to be ready by the end of this year. Several other players have done the task in specific condition as for the SK loop using scenarios.

Using scenarios is not seen in the same way by all players. IMO, if a pattern push you to study say up to 16 scenarios (usually much less), this remains much easier for a player that to explore 200 candidate starts and I prefer to expand dynamically one scenario within an exotic pattern than to look everywhere.

Some clues for the remaining difficulty:

- usually, puzzles having a double exocet collapse or have a remaining very low rating (less than 8)
- rank 0 logic leaves you, usually with a rating around 9 starting from 11
- sk loop never leaves you over 9 unless you don't use the sk loop properties in the next steps.
- a mix of sk loop + exocet or exocet + rank0 is always easy to solve
- Symmetry of given as a double exocet, collapses or is easy to solve