Mike Hansen wrote:
> Hello,
>
> While looking at Maxima 5.14.0 for inclusion in Sage, we notice the
> following bug.
>
> (%i1) limit( %e^x/x, x, inf);
> (%o1) inf
> (%i2) limit( -%e^x/x, x, inf);
> (%o2) inf
>
>
Oops.
> Also, there were the following regressions:
>
> In 5.13.0
> (%i3) integrate(log(1-x^2)/x, x);
> (%o3) log(x)*log(1-x^2)+li[2](1-x^2)/2
>
> In 5.14.0
> (%i31) integrate(log(1-x^2)/x, x);
> (%o31) 1*log(x)*log(1-x^2)-1*((-2)*(-log(1-x^2)/2)*log(x)
> -2*(log(x)*log(1-x^2)+li[2](1-x^2)/2)/2)
>
>
expand(%) makes it the same as 5.13.0, so it's not actually wrong. But
that it's not simplified (note the multiplication by one), is not good.
> In 5.13.0:
> (%i9) bfhzeta(1.1,1/2,50);
> Warning: Float to bigfloat conversion of 12.103813495683745
> (%o9) 1.2103813495683744469025853545548130581952676591199b1
>
> In 5.14.0:
> (%i32) bfhzeta(1.1,1/2,50);
> (%o32) 1.210381349568374531600056798197329044342041015625b1
>
>
I'm not sure about this one, but I think 5.14 is right. Previously,
conversion of float's to bfloat's went through a process of
rationalization, which is not right. In 5.14, floats are converted to
bfloats by, essentially, appending enough zero bits to the float to make
it a bfloat. In 5.13, the nearest rational to the float was computed,
and the rational was converted to a bfloat.
Ray