[Jprogramming] puzzle

[Leo Vo~handu]

>Even if you give additionally also diagonalwise sums that does not give a
unique solution.
In some old Comm ACM there was a study which did show that
01  10
10  01  type uncertainties will stay anyhow.

Leo Vohandu

 The answer is no, though it might be true for small enough arrays (I
> assume you specifically have only 0,1 entries). A problem like this has
> been used in ecology to study bird species distribution. In one case +/q
> and +/"1 q are vectors, one of length 56, the other length 28, The
> number of possible solutions is in excess of 10^100.
>
> There are different algorithms depending on whether you start with a
> known solution or just the vectors (if you want them I can supplu boyj).
>
> Ralph Selfridge
>
> On Thu, 10 Nov 2005, Eugene McDonnell wrote:
>
>> Is it possible to derive q accurately from +/q and +/"1 q?
>>
>>     q=: ?. 5 5 $ 2
>>     q
>> 0 1 1 0 0
>> 1 0 1 0 0
>> 0 0 1 0 1
>> 0 0 0 0 0
>> 1 0 0 0 0
>>     +/q
>> 2 1 3 0 1
>>     +/"1 q
>> 2 2 2 0 1
>>
>> Eugene
>> ----------------------------------------------------------------------
>> For information about J forums see http://www.jsoftware.com/forums.htm
>>
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm