Numerics in C++

[nguyenxuanson]

yes, I have installed CDT, but have not set up Cygwin. I am looking for step by step set up intruction
Son

[Khoa Tran]

I'm sorry for not having spare time to answer earlier.

A step by step instruction is probably impossible for me (at least for now). Installing Cygwin is pretty straight forward possibly except choosing which packages to include.

If you're to program in CDT, you need to have make, gcc, g++ under the devel category. If you want to use gsl, it's available under the libs category. If you want to have graphical outputs then you need to have xwindow packages (they are all under X11). Depending on what you do, you may need additional packages.

This link has a list of all packages together with brief explanations.

Unless you change the default path, Eclipse should call the compiler/linker appropriately.

[nguyenxuanson]

cool, I have europa release and it work witg cygwin flawlessly.
open question
just wonder how to have it work other compiler like intel c++ or visual c++.

[Khoa Tran]

I haven't tried but this should work:

By default, Eclipse generates the makefile automatically. In the project properties, deselect this feature. Create a makefile yourself in the project directory. In the makefile, you can call whatever you want, even Solitaire instead of a C++ compiler.

[Khoa Tran]

References for numerics programming in C++
My friend, Dag Lindbo at NADA (KTH), has written some guidelines for numerical C++, short but quite useful. Here's the link: http://www.nada.kth.se/~dag/index.php?n=Dl.GuidelinesForNumericalC

His page also links to a "C++ Programming for Scientists" course and to Object-Oriented Numerics Page (but in scientific computing research, abusing OOP is many times discouraged due to computational inefficiency <- this is still debatable though).


For plotting
Yet another guide on how to do simple 2DPlot (http://na37.nada.kth.se/mediawiki/in...le_2D_plotting) or 3DPlot (http://na37.nada.kth.se/mediawiki/index.php/Simple_surface_plotting), also written also by NADA guys.


Numerical Linear Algebra Library
For matrix/vector operations:

• uBLAS: http://na37.nada.kth.se/mediawiki/index.php/Using_uBLAS. They said that it's "better" than cBLAS (employed by GSL).
• cBLAS (i.e. BLAS in C):

For solving linear systems:

• Dense matrices: LAPACK is the standard
• Sparse matrices: implement an iterative method ourselves Well, there are some library for sparse systems out there but I haven't seen enough testimonials. Moreover, iterative methods are simple enough to implement.

General numerical routines (bessel function, fast fourier transform, random number generation, etc.): GSL is my choice