[Jprogramming] Overflow, underflow, floats

[Erling Hellenäs]

Hi all !
16 digits of precision and a range between 10^306 and 10^_306 is of 
course quite much. Even if problems in theory exists, maybe in practice 
they don't, or they are rare enough for most applications. I would 
believe that people who are responsible for really important 
calculations are also able to assure that their results are correct. 
There are of course ways to estimate the errors and to do calculations 
in ways that minimizes them. Anyway, I think its important to think 
about these things now and then if you work with software development.
Cheers,
Erling

On 2017-09-21 16:39, Henry Rich wrote:
> I don't know of any "random" results.  There are inexact results 
> stemming from the approximations inherent in limited-precision 
> floating-point arithmetic.  If those approximations cause problems in 
> your application, stay away from floating-point.
>
> J does not conform to IEEE always; in particular 0^0 is 1.  We think 
> that is an improvement on IEEE-754.
>
> Henry Rich
>
>
> On 9/21/2017 9:54 AM, Erling Hellenäs wrote:
>> We had problems with random results in another thread, so I looked at 
>> this. It seems obvious that these problems exist. If you want you can 
>> deny it of course. /Erling
>>
>>
>> Den 2017-09-21 kl. 15:44, skrev Raul Miller:
>>> I would start by better understanding the specific problem I was
>>> trying to solve.
>>>
>>> (As opposed to coming up with a solution and then looking for a
>>> problem that fits to it...)
>>>
>>> Thanks,
>>>
>>
>> ----------------------------------------------------------------------
>> For information about J forums see http://www.jsoftware.com/forums.htm
>
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm