floating point inaccuracies when compiling

[Neil Ferguson]

This is the behaviour I'd expect.

The 1.0e-10 is being represented as a double-precision floating point value. 
Not all values can be represented exactly, and in this case the machine picks 
the closest value to 1.0e-10 that it can represent.

Double precision generally provides 15 significant figures of precision; as 
you can see, the value chosen by the machine is precise to that point.

If you really need your computations to have greater precision, in spite of 
any performance penalty, you might like to consider the GMP library:

   http://www.swox.com/gmp/

Neil.

Florian Hackenberger wrote:
> Hi!
> 
> I'm using gcc 3.4.3-20050110 compiled with flags "-O2 -march=i686 -pipe" for C 
> and C++. When compiling expressions like the following
> 
> Array2D< double > numerical_damper = pow(local_model.sum_e_upd2 / 
> (local_model.sum_e_cv2 + 1.0e-10 ), 4);
> 
> I noticed, that the values used for 1.0e-10 are not really 1.0e-10. To 
> visualise that, I forced gcc to produce an error message:
> 
> ./rfwr.cpp: In member function `void Rfwr::updateCovTriangular(const 
> TNT::Array2D<double>&, const TNT::Array2D<double>&, LocalModel&, const 
> TNT::Array2D<double>&)':
> ./rfwr.cpp:251: error: no match for 'operator+' in 
> 'local_model->LocalModel::sum_e_cv2 + 
> 1.0000000000000000364321973154977415791655470656e-10'
> 
> Is this behaviour as expected, or is there something wrong with my 
> installation?
> 
> Thanks in advance!
> 
> Regards,
> 	Florian
>