Comments on Bitwiese: The Joys Of Floating Point Numbers

"This is usually not such a big deal in practice (...

2007-04-12T01:33:00.000+01:00

"This is usually not such a big deal in practice (i.e. scientific computing)."

In computational geometry and solid modeling, floating point problems can be a very big deal.

This is usually not such a big deal in practice (i...

2007-04-12T00:51:00.000+01:00

This is usually not such a big deal in practice (i.e. scientific computing).

We use floating point numbers in numerical algorithms, and usually the algorithmic error (i.e. the error obtained when doing computation over the real numbers) is much larger than floating point error.

I.e., use RK4/5 to solve a differential equation, with dt=0.01. RK4/5 is O(dt^6)=1e-12. Floating point

cshesse, thank you for your comment.

Yes, <...

2007-04-12T00:38:00.000+01:00

cshesse, thank you for your comment.

Yes, arbitrary-precision arithmetics can solve floating point related problems in some cases (e.g. correct financial calculations).

However, you do not profit from the native - and fast - floating point operations of modern processors.

Additionally, some problems are

The python decimal module lets you mess around with arbitrary precision numbers. There is even a (super slow) implementation of the standard math functions for decimal numbers called dmath.

Comments on Bitwiese: The Joys Of Floating Point Numbers

"This is usually not such a big deal in practice (...

This is usually not such a big deal in practice (i...

cshesse, thank you for your comment.Yes, <...

The python decimal module lets you mess around with arbitrary precision numbers. There is even a (super slow) implementation of the standard math functions for decimal numbers called dmath.

cshesse, thank you for your comment.

Yes, <...