# Re: Possreps and numeric types

From: Bob Badour <bbadour_at_pei.sympatico.ca>

Date: Thu, 29 Mar 2007 20:36:39 GMT

Message-ID: <rrVOh.17061$PV3.175756_at_ursa-nb00s0.nbnet.nb.ca>

Date: Thu, 29 Mar 2007 20:36:39 GMT

Message-ID: <rrVOh.17061$PV3.175756_at_ursa-nb00s0.nbnet.nb.ca>

Gene Wirchenko wrote:

*> Bob Badour <bbadour_at_pei.sympatico.ca> wrote:
**>
**>
*

>>Gene Wirchenko wrote: >> >> >>>Bob Badour <bbadour_at_pei.sympatico.ca> wrote: >>> >>> >>>>Gene Wirchenko wrote: >>>> >>>> >>>>>Bob Badour <bbadour_at_pei.sympatico.ca> wrote: >>>>> >>>>> >>>>>>Gene Wirchenko wrote: >>>>>> >>>>>> >>>>>>>Bob Badour <bbadour_at_pei.sympatico.ca> wrote: >>>>>>> >>>>>>>[snip] >>>>>>> >>>>>>> >>>>>>>>I admit I used epsilon somewhat sloppily and not necessarily with the >>>>>>>>exact meaning used when discussing a particular floating-point >>>>>>>>implementation. I used it to mean the distance to the representable

>

> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

*>*>>>>>>>>predecessor or successor of any representable rational value.

>

> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

*>*>>>>>>> That is incorrect. It is per >>>>>>> http://en.wikipedia.org/wiki/Machine_epsilon >>>>>>>In a floating-point system, epsilon is the smallest number such that

>

> ^^^^^^^^^^^^^^^^^^^^^^^^^

>

>>>>>>> 1 + epsilon > 1

>

> ^^^^^^^^^^^^^^^

*>*>>>>>>>[snip] >>>>>> >>>>>>When you say the above is incorrect, are you saying I was not using >>>>>>epsilon sloppily? Or are you saying I didn't use it to mean the distance >>>>>>to the representable predecessor or successor of any representable >>>>>>rational value? >>>>> >>>>> The definition of epsilon. >>>> >>>>Other than the "scaled by the exponent" bit, what exactly is the >>>>difference? ie. If I subtracted one from the other, what answer would I get? >>> >>> If I am understanding you, other than that difference, none. The >>>answer would be garbage. >> >>I really don't understand your objection. I noted I used the term >>sloppily and not as defined for floating point when I was discussing >>rational representations. Using the only definition I gave for the >>floating point version, the difference would be zero. The definitions >>are equivalent only stated differently.

>

> No. The definition that I gave is in terms of 1, not just any

*> representatable value.*I guess I will just roll my eyes and move along. Received on Thu Mar 29 2007 - 22:36:39 CEST