Le forum de XCAS

Consider the following code
The above returns 0, but it must return a normalized f (f divided by the coefficient of the monomial with highest degree). Any ideas how to resolve this?

Jose

This is not giacpy related, I have a similar problem with n>=12:
whatever is epsilon or Digits. (I put evalf with 100 digits to avoid global Digits or epsilon2zero limitations. Bernard, is there some epsilon2zero use in this division? or is it always done in floats instead of mpfr?

Frederic

Yes, there are eps->0 replacements. I have removed one in regroup in my source. But I won't make a lot of changes related to that, because I fear side effects.

frederic han a écrit : ↑

ven. févr. 15, 2019 6:02 pm

This is not giacpy related, I have a similar problem with n>=12:
whatever is epsilon or Digits.

Code : Tout sélectionner

n:=12:; expand((evalf(10**n,100)*x^2+x)/evalf(10**n,100))

(I put evalf with 100 digits to avoid global Digits or epsilon2zero limitations. Bernard, is there some epsilon2zero use in this division? or is it always done in floats instead of mpfr?

Frederic

Thanks for the explanation.

That's a pity. In your example n=12 is rather small and I think changing Digits will have no effect when using giacpy in this way. As you probably know, XCas has the correct result . So is it correct to assume that as soon as I have a division of this sort with python and giacpy I am using floats and not mpfr/gmp or is it that the global epsilon is set much lower? In my XCas epsilon is set to 1e-12 and giacpy claims the same (if I type giac("epsilon") I get 1e-12). I suspect the epsilon is higher in practice (when using giac outside XCas) because your example for instance is still within float precision (I think).

So I am guessing there is no workaround for this (even a hack). This is a pity since I am solving some system of equations for which I would need the numerical values.

Jose

No my previous example gives 0 under Xcas also.

Digits is avaible with giacpy see: