I propose a test for generator 3 in the PADLS experiment.

Generator 3 generates a lot of pseudo-associative DLSs.
I tested for 500,000 DLSs.
Found 283776 CF and 3 ODLS (--> 6 CF ODLS).

Поиск марьяжных ДЛК (кроме симметричных) для семейства ЛК

Введено ЛК: 283776

Найдено марьяжных ДЛК: 3 они записаны в файл output.txt
Время работы в сек   : 16497.9

In the proposed test introduced to generate 2 million DLSs.

Please perform this test and report the results.

The row to generate is already written to the file A18.txt.
Run the gener_check.bat script.

Link
https://cloud.mail.ru/public/CEF7/4zU8xhrGZ

Natalia Makarova, It is possible to include the third generator, but I am not ready yet. What I can do, is load another batch for the current application, if you want. I do not know how serious is the generator error with rows starting with 7 and 8. If you want to send me a next batch, then please make sure the existing error does not affect solutions. I can send you the raw results soon. Processed results after a batch is finished.

Tomas Brada
you understood me wrong.
I proposed a test for generator 3 in manual mode, and not in the BOINC project.

I still do not suggest you include in the BOINC-project.

The other batch of WUs, as I already wrote, for modified generators 1 and 2.
These versions of the generators must be redone for the BOINC project.

But first you need to complete the testing of the existing two batches and process all the results obtained.

The bug with rows starting with 7 and 8 has already been fixed. This mistake was only in batch 2.
In the following batches there will be no such error, because I have corrected the error in the rows generation program.
[The rows starting with 7 and 8 in batch 2 are not processed and do not produce solutions!]

PS. I am in no way rush you!

By the way, WUs generation for generator 3 will not be performed by the client, because this generator produced too many pseudo-associative DLSs for one row.

This is the first pseudo-associative DLS produced by generator 3



We have a violation of associativity in 6 central rows.
This is the lowest degree of associativity, compared with generators 1 and 2.
Therefore, there are a lot of such pseudo-associative DLSs.

XAVER perform the test.

In the test task, there are 2 million pseudo-associative DLSs.
The test was performed 7.5 hours.
Found 4 ODLS

0 3 4 2 7 9 8 5 6 1
6 1 7 9 2 3 4 8 0 5
5 7 2 8 1 4 0 3 9 6
8 9 0 3 6 7 2 1 5 4
3 5 8 1 4 6 9 0 2 7
1 2 6 7 8 5 3 9 4 0
9 8 1 5 3 0 6 4 7 2
4 0 9 6 5 2 1 7 3 8
2 4 5 0 9 1 7 6 8 3
7 6 3 4 0 8 5 2 1 9

0 4 6 9 8 7 2 5 3 1
5 1 3 0 9 4 7 8 2 6
1 5 2 4 6 0 9 3 7 8
9 7 0 3 2 8 4 1 6 5
3 2 5 8 4 6 1 9 0 7
6 9 8 2 7 5 3 0 1 4
7 8 1 5 0 2 6 4 9 3
2 3 9 6 5 1 8 7 4 0
4 0 7 1 3 9 5 6 8 2
8 6 4 7 1 3 0 2 5 9

0 9 4 2 8 3 7 5 6 1
6 1 7 4 0 8 9 3 2 5
9 7 2 6 3 4 1 0 5 8
5 0 8 3 9 1 4 6 7 2
8 2 5 1 4 6 0 9 3 7
1 4 6 9 7 5 8 2 0 3
2 3 0 8 5 7 6 1 9 4
4 8 9 5 6 2 3 7 1 0
7 5 3 0 1 9 2 4 8 6
3 6 1 7 2 0 5 8 4 9

0 9 4 7 6 3 8 2 5 1
3 1 7 4 8 2 5 9 0 6
8 7 2 9 1 4 0 3 6 5
5 8 6 3 9 7 4 0 1 2
1 3 0 2 4 6 7 5 9 8
9 0 1 6 7 5 2 8 3 4
7 2 5 8 3 9 6 1 4 0
4 6 9 5 0 8 1 7 2 3
6 5 3 1 2 0 9 4 8 7
2 4 8 0 5 1 3 6 7 9

The first three solutions I found in the test for 500,000 pseudo-associative DLSs.
Additionally found one solution.
All solutions are unique!

Characteristics of the generator 3:

1. generates a lot of pseudo-associative DLSs;
2. solutions gives less than generators 1 and 2;
3. higher percentage of unique solutions.

I have a problem with generation. I can not generate all the DLSs for a given row, because there are very many of them.
I showed tests for 500,000 DLSs and 2,000,000 DLSs.
But DLSs is actually more.

XAVER
thank very much for testing!

Having 2'000'000 and more squares to test in one workunit is getting out of the comfortable range. It will still work, do not get me wrong. But perhaps a 9-th parameter should be added to the program that will restrict the generator to only a portion of the space. For example in the second generator there is a outer loop:

FOR I1=2 TO 9

. Instead of looping from 2 to 9, a parameter should select which value of I1 to check.

Can this generator produce the same one square when started with different row numbers? This should be prevented if possible.
Can this generator produce square which the first or second one produced?
How many squares with symmetry class that we have generators (eg 4,31,31) are produced? Would it be viable to eliminate such squares before starting the check with family_mar?

Having 2'000'000 and more squares to test in one workunit is getting out of the comfortable range. It will still work, do not get me wrong. But perhaps a 9-th parameter should be added to the program that will restrict the generator to only a portion of the space.

I understand what you are saying.
I already have more than one generator (in other experiments!), which produces many millions of LS.
Harry White copes well with this. He knows how to stop the generation for a given number of squares and remember the last generated square. The next time you start the program will start the generation from this square.
I do not know how to do it.

Can this generator produce the same one square when started with different row numbers? This should be prevented if possible.

It is excluded.

Can this generator produce square which the first or second one produced?

I tried to exclude it.
However, some intersections are inevitable due to the DLS isomorphism.

How many squares with symmetry class that we have generators (eg 4,31,31) are produced? Would it be viable to eliminate such squares before starting the check with family_mar?

In the PDALS experiment, solutions with symmetry are rarely found (4,31,31).
This is unpredictable and impossible to exclude.

Generator 3 covers tens of times more pseudo-associative DLS space.
This variant cannot be performed manually.

However, there is still a complete space of pseudo-associative DLS!

I am already testing software for this branch of the PADLS experiment.

Having 2'000'000 and more squares to test in one workunit is getting out of the comfortable range.

Tomas Brada
In the current experiment, you divide even smaller WUs :)