Re: Re: Re: Bug in 5.1??
- To: mathgroup at smc.vnet.net
- Subject: [mg54578] Re: [mg54528] Re: [mg54469] Re: Bug in 5.1??
- From: DrBob <drbob at bigfoot.com>
- Date: Wed, 23 Feb 2005 03:11:56 -0500 (EST)
- References: <200502210844.DAA27157@smc.vnet.net> <200502220923.EAA10793@smc.vnet.net> <03d46b8e970e9ccc74c8324d28644629@mimuw.edu.pl> <opsmlt5wcgiz9bcq@monster.ma.dl.cox.net> <d279073ca556051f048e823d9cfcb42e@mimuw.edu.pl>
- Reply-to: drbob at bigfoot.com
- Sender: owner-wri-mathgroup at wolfram.com
>> I, on the other hand, am telling him that
>> what he has observed has been well known for 10 years +
It hasn't been well known to ME, to the original poster, or to Dana DeLouis. Or, I think, to a lot of other people.
I did know about some other issues, including spurious loss of precision in certain Newton iterations, but I didn't know that N[number,2] doesn't have two digits of precision.
Anyway, I think we're entitled to discuss well known issues, if we want. If nobody's interested, a thread dies quickly.
If the word "bug" offends you, sorry 'bout that. Many of us use the word somewhat loosely, I suppose, but I'm not aware of any reliable way to determine what does or doesn't deserve the name.
Bobby
On Tue, 22 Feb 2005 16:35:52 +0100, Andrzej Kozlowski <akozlowski at gmail.com> wrote:
>
> On 22 Feb 2005, at 16:18, DrBob wrote:
>
>> I don't file bug reports because (a) I don't usually have real
>> applications that warrant it (certainly not in this case) and (b)
>> reporting bugs is a pretty fruitless endeavor -- even if WRI _is_ a
>> hundred times more responsive than Microsoft (for instance) in that
>> regard. Nothing gets fixed until the next release, and many bugs have
>> already survived five years or more. I am likely to get a work-around
>> from tech support, but I can come up with work-arounds on my own.
>>
>> Anyway, WRI can't claim to be unaware of this problem, or anything
>> else we discuss in this forum -- regardless of whether anybody files a
>> bug report.
>>
>> I do understand that developmental priorities and theoretical issues
>> make it impractical to fix some of these things. Generally, I'm not
>> really asking anyone to fix them; I just want to understand, to
>> whatever extent I can. Possibly you think I'm far more concerned than
>> I actually am.
>>
>> As for 2-digit significance arithmetic in particular, if I want an
>> explanation and you don't have one, and you clearly don't care about
>> the issue, why answer?
>>
> Because when you tell anther user that he has "discovered" a bug, you
> are implying that he should take the trouble to report it, which is
> something you do not wish to do yourself because you know well that it
> will be a waste of effort. I, on the other hand, am telling him that
> what he has observed has been well known for 10 years + and nothing
> has been done about it, for a variety of possible reasons, the most
> likley being that this "problem" is considered harmless; other
> possible reasons being that nobody knows how to fix it or whether
> fixing it would damage other things or even perhaps because
> significance arithetic can't work without something like that.
>
> Andrzej Kozlowski
>
>
>
>
>
>>
>> Bobby
>>
>> On Tue, 22 Feb 2005 15:37:35 +0100, Andrzej Kozlowski
>> <akozlowski at gmail.com> wrote:
>>
>>> Your response is very curious. Who told you you are not supposed to
>>> complain? I suggested you file a bug report. If you do not file one
>>> presumably you have a reason?? Is the reason that (you imagine) I told
>>> you not to complain? I am flattered to see that I now wield this sort
>>> of authority but really I do not seek it. I only told you that in the
>>> past people wrote write about such issues and, in the old days,
>>> someone
>>> (usually David Withoff) would respond. But even he never gave, as far
>>> as I can recall, a clear reason why this happens, only a vague one
>>> like the one I have given: strange things happen if you use
>>> significance arithmetic at low precision,and try making comparisons
>>> like
>>>
>>> 10`2 ==0
>>>
>>> True
>>>
>>> or
>>>
>>> 100`2==0
>>>
>>> True
>>>
>>> or whatever. Don't do it.
>>>
>>> So now please do file a bug report and demand an explanation. I will
>>> be
>>> very interested to read the response, if you get one of course. After
>>> all, I am just another user and have no better access to WRI's
>>> thinking
>>> than you have.
>>>
>>> Andrzej Kozlowski
>>>
>>>
>>>
>>>
>>> On 22 Feb 2005, at 10:23, DrBob wrote:
>>>
>>>> It doesn't make sense to ANYBODY, as far as I can tell.
>>>>
>>>> But we're not supposed to complain, because somebody ALREADY
>>>> complained (some time or another).
>>>>
>>>> Bobby
>>>>
>>>> On Mon, 21 Feb 2005 03:44:38 -0500 (EST), Dana DeLouis
>>>> <delouis at bellsouth.net> wrote:
>>>>
>>>>> Seems to me that the loss of precision returns an "Interval", which
>>>>> is still
>>>>> greater than zero. It looks like the cutoff between True & False is
>>>>> a hair
>>>>> above -1, which doesn't make sense to me either.
>>>>>
>>>>> Interval[N[5, 2]]
>>>>> Interval[{4.9, 5.1}]
>>>>>
>>>>> N[5, 2] > -1
>>>>> True
>>>>>
>>>>> N[5, 2] > -1 + $MachineEpsilon
>>>>> False
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> DrBob at bigfoot.com
>>>> www.eclecticdreams.net
>>>>
>>>>
>>>
>>>
>>>
>>>
>>
>>
>>
>> --
>> DrBob at bigfoot.com
>> www.eclecticdreams.net
>>
>
>
>
>
--
DrBob at bigfoot.com
www.eclecticdreams.net
- References:
- Re: Bug in 5.1??
- From: "Dana DeLouis" <delouis@bellsouth.net>
- Re: Re: Bug in 5.1??
- From: DrBob <drbob@bigfoot.com>
- Re: Bug in 5.1??