Monday, October 11, 1999 Volume 99 : Issue 40
From: Martin Berzins
Date: Mon, 27 Sep 1999 20:11:03 +0100
Subject: Is Scientific Computing Part of Computer Science?
Is Scientific Computing part of Computer Science?
Every 5 years or so in the UK there is a Research Assessment Exercise
whereby the research of all University departments is examined.
The definition of Computer Science for the forthcoming exercise makes
no mention of Scientific Computing. I'd be interested in the reaction
of the NA-digest community to this.
From: Ken Turkowski
C T H Baker
Date: Tue, 5 Oct 1999 08:37:59 -0700
Subject: Re: Is Scientific Computing Part of Computer Science?
In last week's NA Digest, Martin Berzins wrote:
> Every 5 years or so in the UK there is a Research Assessment Exercise
> whereby the research of all University departments is examined.
> The definition of Computer Science for the forthcoming exercise makes
> no mention of Scientific Computing. I'd be interested in the reaction
> of the NA-digest community to this.
Ken Turkowski replies:
Similarly, I'm amazed that very few computer scientists know anything
- frequency analysis and filtering (FFT, convolution)
- numerical analysis
Several times annually, I see someone inverting a matrix by computing
cofactors and determinants (accumulating in single-precision!). This sort
of naivete has got to be corrected!
Immersive Imaging Technologist
Apple Computer, Inc.
Bill Silvert replies:
This is an interesting question, but should be broken into two parts --
SHOULD it be a part, and IS it a part. Clearly the needs of scientific
computing are of academic interest, and some computer scientists work in
this area, but many departments overlook it.
The case of FORTRAN illustrates this. FORTRAN is widely used for
scientific computing, but many CS departments refuse to teach it on the
grounds that it is poorly constructed and is not a good subject for
teaching CS. True, but this gap between theory and application has some
pretty negative side effects.
Habitat Ecology Section
Bedford Institute of Oceanography,
Dartmouth, Nova Scotia, CANADA
Joe Grcar replies:
Computer science began as an interdisciplinary subject
once synonymous with scientific computing. Today, the
two have little in common besides the transposed names.
The truth of this is revealed by a cursory inspection of
major computer science journals. There are few if any
articles about scientific computing. Examine, for example,
the annual percent of articles on any numerical subject
in JACM, the flagship journal of the ACM. The numerical
content peaks at over 60% in 1960, and fluctuates well
below 10% since 1980.
The reason is, scientific computing remains inter-
disciplinary. The question it addresses is numerical
calculation. This is important but merely procedural
to the sciences; it is not the goal of their intellectual
pursuit. Moreover, as Steve Smale pointed out in his
SIAM von Neumann lecture, numerical analysts never
discovered a theoretical framework that simultaneously
treats questions of error propagation and algorithmic
complexity. So when computer science became an
established academic subject in the 1960's and 70's,
the intellectual core formed around the logical and
combinatorial theory of algorithms without accommodating
questions of numerical error. The latter can be viewed
as part of the mathematical theory of approximation, so
I would expect numerical analysis to find a place there.
Thus, responding to Martin Berzins' question in NA Digest
(v 99 n 39), "is scientific computing part of computer
science?", the answer is "no". We might like the answer
to be different, but the facts plainly indicate it is not.
The point that Martin should make to the authorities
is something like this. Scientific computing is an
integral part of industrial design and scientific research.
One can even argue it is an important aid to making
government policy, since scientific computing is used
to predict the future climate and to develop weapons.
But scientific computing done well is an interdisciplinary
undertaking. None of the participating fields can claim
they alone do the whole thing well, or that projects
in the interdisciplinary subject can compete successfully
for a given field's research funds.
So, given that research in scientific computing is important
and that it is not part of any established program, then
how do the authorities propose to support it?
Lacking some answer to this question, I expect the current
Balkanization of scientific computing to continue. The
inevitable outcome of this will be the use of suboptimal
computing methods and the unrealized potential to address
important scientific and engineering problems.