[Date Index] [Thread Index] [Author Index]
Re: Mathematica Weirdness
On 2/25/11 at 6:32 AM, sebhofer at gmail.com (Sebastian) wrote: >On Feb 24, 12:26 pm, Daniel Lichtblau <d... at wolfram.com> wrote: > >>If instead you numerically evaluate using precision tracking you >>will get a positive result. >I'm curious: why does N[#,MachinePrecisison] evaluate to the wrong >result while N[#,n], where n is any number (also smaller than 16), >evaluates to the correct result? What is so special about >MachinePrecision? When using MachinePrecision, floating point operations are done by your FPU and are dependent on the specifics of your platform. Using anything other than MachinePrecision invokes Mathematica's arbitrary precision algorithms and it is Mathematica that controls the floating point operations rather than simply handing things over to the FPU. The advantage of Mathematica algorithms over hardware (FPU) can be a better result. The cost is slower execution.