Thread: A numerical problem in matlab

In matlab,

ezplot('log(1+x)/x',[-10^(-15),10^(-15)])

and

ezplot('log(1+x)/(1+x-1)',[-10^(-15),10^(-15)])


will give different plot. Why? I know there may be because of the representation erro of the range -10^(-15). But I don't know exactly. Could anyone tell? Thanks a lot!!

Originally Posted by seekor

In matlab,

ezplot('log(1+x)/x',[-10^(-15),10^(-15)])

and

ezplot('log(1+x)/(1+x-1)',[-10^(-15),10^(-15)])


will give different plot. Why? I know there may be because of the representation erro of the range -10^(-15). But I don't know exactly. Could anyone tell? Thanks a lot!!

Floating point arithmetic is non associative, so x != 1+x-1, this is
particularly pronounced near the limit of the prescission of a double
(which is the Matlab default floating type)

RonL

Originally Posted by CaptainBlack

Floating point arithmetic is non associative, so x != 1+x-1, this is
particularly pronounced near the limit of the prescission of a double
(which is the Matlab default floating type)

RonL

Thanks for your attention! RonL.

However, I am still not very clear why 1+x-1 can give better result than x. I thought that when x is very small, then 1+x will become 1 and then 1+x-1 will become 0....