Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cornell!uw-beaver!blake!keffer
From: keffer@blake.acs.washington.edu (Thomas Keffer)
Newsgroups: comp.lang.c++
Subject: Vector and Matrix classes
Keywords: C++, matrix, vector
Message-ID: <1878@blake.acs.washington.edu>
Date: 4 May 89 18:03:21 GMT
Reply-To: keffer@ocean.washington.edu (Thomas Keffer)
Organization: University of Washington, Seattle
Lines: 35

There has been a lot of discussion on implementing a set of vector and
matrix classes in C++, particularly with respect to minimizing
temporary variables.

I am wondering what implementations exist out there?  Keith Gorlen has
supplied a fairly complete set of Vector classes as part of his "NIH
Class Library" (nee OOPS), but these are overly general for many
applications and can be slow because the copy initializer relies on
copying all elements.

What else?  Is this the state of the art?

One can think of many useful extensions besides the obvious arithmetic
operations (+, *, -, etc.) such as:

1) Classes for specialized types of matricies, such as symmetrical or
tridiagonal.  These would use "optimized" algorithms for such
operations as solving linear sets of equations, etc.

2) A class LU_decomposition?  Contains the LU decomposition of a
matrix.  Would include member functions such as invert(), or solve().

3) There HAS been some discussion of FFT algorithms (principally from
Dr. James Coggins and myself).  Any implementations?  In particular,
what other algorithm servers have been implemented?  

I have made a start in these three areas by building specialized
derived classes on top of the NIH classes.  But they can become
unwieldy because of all the underlying machinery.  I was hoping to do
something cleaner and build these specialized operations on top of a
leaner set of matrix and vector classes.

But, I'm coming to appreciate that for all their bulk and sometimes
overgenerality, the NIH classes have the overwhelming advantage that
THEY EXIST!!!