%% This document created by Scientific Word (R) Version 2.0
\documentstyle[path,twocolumn]{article}
\title{Netlib News: BibNet augments preprint servers and electronic journals}
\author{Eric Grosse}
\date{30 Jan 1995}
\begin{document}\maketitle

Mathematics is rapidly moving toward electronic publishing, but no one knows
quite what shape it will take in our community. Will we use a centrally
managed archive, analogous to netlib for math software and the eprint server
at Los Alamos for physics papers? Will we publish in new journals like ETNA
or electronic forms of established journals like ESISC? Will we depend on
PostScript ftp archives at each research lab? All the above?

One thing is clear---locating a particular paper amidst this chaos can be a
challenge. BibNet, a new service operated by Nelson Beebe and Stefano
Foresti and hosted by netlib, offers an appealing answer to this challenge.
Readers search this database or
download individual author bibliographies by directing your favorite
Web browser to \path|http://netlib.att.com/netlib/bibnet.html|
or by a request like
\emph{find gold in bibnet} mailed to netlib.

The compelling idea behind BibNet is that authors have the motivation to
keep complete, accurate, and timely bibliographic information about their
own work. No one else is as well positioned to keep track of when their
preprints appear, or when those preprints turn into final publications.
Authors already have to maintain this information in a curriculum vitae for
professional reasons, so there is little extra work for them.

BibNet has chosen to standardize on BibTeX, already widely adopted in math
typesetting, as the format for these bibliographies. (If you're still using
refer format, just ask netlib to \emph{send bibnet/tools/software/r2b}
for a converter.) A description can be found in your
\LaTeX\ manual or in the following:
\begin{verbatim}
  @Article{Higham:1994:BVT,
  author="Nicholas J. Higham",
  year=1994,month=January,
  title="{\BibTeX}: A Versatile
  Tool for {\LaTeX} Users",
  journal="SIAM News",
  volume=27,number=1,
  pages="10,11,19"},
  URL="ftp://vtx.ma.man.ac.uk/pub/higham/bibtex.dvi.Z"
\end{verbatim}

Each entry contains a key that BibNet makes up from the first
author, the year, and initials from the title.
This is guaranteed to be unique across the entire
database and unchanging across time. Most BibTeX fields such as month are
optional, but even the flimsiest preprint has to have an explicit author,
year, and title.

Authors in scientific computing are invited to send their publication list
to \emph{bibnet-submit@math.utah.edu}.
When you first send your bibliography, include whatever contact information
you wish to have listed at the beginning of your bibliography. Customarily
this includes name, affiliation, paper address, and email address. After
some checking of the material, you will receive an a acknowledgement and
instructions on sending updates.
More information on BibNet
and instructions for contributors can be found in the readme file
in the top level directory of BibNet, or by sending email
to \emph{bibnet-info@math.utah.edu}.  

As mentioned at the beginning, BibNet lives in the context of an exciting
change in mathematical communication. An extra URL field is encouraged in
bibliographic entries, pointing to the online form of the paper. For
mathematics, this will probably best be a compressed PostScript file on an
ftp or http server at the author's institution.

Sign up now, and get your work before a wider audience!



\section*{Recent additions to netlib}

People interested in how C++ might affect the structure of scientific
programs will want to look at \path|netlib/diffpack| by
SINTEF and the University of Oslo.  From the documentation:
``Diffpack consists of a collection of object-oriented libraries for
solving partial differential equations, and several Unix utilities for
general software management and numerical programming. In particular,
this piece of software is aimed at rapid prototyping of simulators based
on PDEs, still offering a high level of efficiency.''

A different perspective on rapid numerical prototyping is offered by
\path|netlib/env/yorick/.tar.gz|, which defines
an interpreted language like Lisp, but with C-like syntax, an
emphasis on arrays, interactive X graphics, a library of functions
(Bessel, least squares fitting, netCDF, etc.) written in the Yorick language,
and a mechanism for linking compiled functions.

The contour plot is one of the most important graphical techniques in
scientific computing, second only to scatter/function plots.
The tool \path|netlib/graphics/xfarbe.taz.uu| by A.~Preusser
provides filled contours, legends, contour labels, extrema labels,
and interactive data probing, producing output on X displays or Postscript.
It uses bicubics on rectangles, so an initial interpolation
stage may be necessary in some applications.  This is a big topic (with
useful programs available from netlib) but for now I'll just note that
there are times when a crude contour plot based on piecewise linear
triangles may be both computationally efficient and best for avoiding
interpolation artifacts.  When a raster image is desired, direct
evaluation at pixels (with antialiasing) works well.
But for most publication and data exploration purposes,
\path|xfarbe| is the best freely-available package around.

A program for computing minimal surfaces using Sobolev gradients
was added to \path|netlib/a/| by R. J. Renka and J. W. Neuberger.

The first release of the Interprocessor Collective Communications Library
(by Barnett, Payne, Gupta, Shuler, van de Geijn, Watts) went into
\path|netlib/intercom|.  If you're using the NX library on the Intel
Paragon, you'll want to look into this.  Or if you're developing an
efficient MPI library for another machine, the authors hope you can
learn some tricks from studying Intercom.

T. Davis and I. Duff offer \path|linalg/umfpack.shar| for
solving sparse unsymmetric linear systems by the multifrontal method.

In the \path|netlib/aicm| library (for the new journal
Advances in Computational Mathematics) 
Marco Marletta contributed codes for solving
eigenvalue problems for Hamiltonian systems.

The Numerical Algorithms journal published, in directory \path|numeralgo|:
\path|na5|: near-breakdown in CGS, by Brezinski and Zaglia;
\path|na6|: DQAINF - integration of infinite oscillating tails,
by Espelid and Overholt.

Recent additions in the \path|toms| directory from the ACM Trans.~on
Math.~Software are:
\path|731|: adaptive moving grid for univariate partial differential equation,
by J. G. Blom and P. A. Zegeling;
\path|732|: nonseparable self-adjoint elliptic PDE on 2D polygonal domain via
capacitance matrix, Laplacian preconditioner, FACR,
by P. F. Cummins and G. K. Vallis;
\path|733|: optimal control problem, by D. Kraft;
\path|734|: translation of Algorithm 630 into Fortran90, by A. G. Buckley;
\path|735|: pyramid wavelet transform and inverse, by C. Taswell and K. C. McGill.

Noteworthy version updates are:
\path|ampl|;
\path|f2c|;
\path|fdlibm| 5.2, 95/01/08;
\path|fp/dtoa.c|;
\path|ftnchek| 2.8.1;
\path|linalg/qmrpack|;
\path|pdes/mgghat|;
\path|pltmg| 7.1;
\path|pvm3| 3.3.6;
\path|research/siamdb.enc| (SIAM membership list) Oct 1994;
\path|slatec| 4.1;
\path|toms/566| remark by Averbukh, Figueroa, Schlick.

The widely admired multiple precision arithmetic package \path|mpfun|
by D. Bailey is now available in both Fortran77 and Fortran90 versions.

\path|ode/cvode.tar.Z| by Cohen and Hindmarsh
combines earlier vode.f and vodpk.f
to solve large non-stiff or stiff ordinary differential
equation initial-value problems.  It is written in C (my favorite
language!)

One of the particularly interesting additions this period
was Leveque's
\path|pde/claw| package for solving hyperbolic systems of
conservation laws in one and two space dimensions.
More about that next time.

{\em Eric Grosse is at the Computing Science Research Center,
AT\&T Bell Laboratories, Murray Hill NJ 07974, USA.
\path|http://netlib.att.com/netlib/att/cs/home/grosse.html|}

\end{document}