XML, why Mma is popular, was...Re: [Maxima] Lunch with George Carrette
Thu, 30 Oct 2003 22:33:15 -0500
At the risk of some repetition, let me attach the history
of Macsyma as I see it. I know the history after 1986 first-
hand, and I have tried to incorporate as much about the
earlier period as I can. Please send me any further
corrections. (I may not have include all that have been
sent to me, for which I apologize.)
Regarding 'marketing glitz', Wolfram found the right way to
exploit the mismanagement of Macsyma by M.I.T. and Symbolics.
He started by focusing on notebook interface and graphics,
and then on numerics, while Mathematica's symbolic math was
still very weak. Wolfram correctly assessed what most people
wanted, and what he should do first before tackling the
extremely demanding task of building a first-class Symbolic
math engine. Given his background in physics, he understood
what scientists wanted, but he did not understand at that
time how to build a great symbolic engine, nor did he
understand what engineers wanted, so that WRI still does not
do so well in that segment. (I now work for MathWorks which
does understand what engineers want, and it is by and large
As you can read below, the failure of Macsyma required the
failure of academic, government, commercial, and philanthropic
Sectors -- quite a stunning record.
Macsyma was developed at M.I.T. from 1968 to 1982 with funds from
the U.S. Defense Advanced Research Projects Agency and some from the
Department of Energy. The government lost interest in Macsyma around
1977 when numerical analysts persuaded the government than numerical
libraries on supercomputers were better suited to performing the
engineering computations needed by the defense establishment.
In the late 1970s the professor in charge of Macsyma at M.I.T. wanted
to lead a company to commercialized Macsyma while he remained a
professor at M.I.T. M.I.T. has never permitted this sort of arrange-
ment, and would not approve the plan. Ever since that time, M.I.T.
has had an ambivalent policy toward the success of Macsyma. (For
example, around 1982, an M.I.T. professor close to Macsyma told
Charles Johnson, the owner of IMSL (the leading numerical software
vendor of the 1970s and 1980s), that Macsyma was obsolete and that
IMSL should not try to commercialize it.)
Richard Fateman worked on Macsyma in the early 1970s and took a
position in the Computer Science department at U.C. Berkeley around
1976(?). He wanted access to the Macsyma sources to continue his
work but M.I.T. denied the request. This was the opening round
in a series of bitter disputes over Macsyma that involved M.I.T.,
the U.S. Department of Energy, U.C. Berkeley, and the U.S. National
Labs, mostly Los Alamos. At that time in the U.S. there were broader
policy disagreements about whether universities could privatize
research results obtained on government research contracts. (By the
end of the 1980s, government policies resolved this issue in the
1.2 Symbolics, Inc.
Symbolics acquired an exclusive license to Macsyma in 1982. The
Symbolics effort did not go well for three reasons. First, Symbolics
was so focused on selling Lisp machines that it gave inadequate funding
to Macsyma and initially impeded the contractually required ports of
Macsyma to other types of workstations. Secondly, the Macsyma group
lacked business leadership and had a lot of internal quarrels. Thirdly,
some government agencies, notably the Department of Energy and the
national labs, refuse to buy from Symbolics, and began pouring
significant resources into an alternative government version of Macsyma,
called 'DOE Macsyma.' At this time, Wolfram's first symbolic product,
SMP, was seriously eating into Macsyma's market, though it was a
markedly inferior product.
I joined Symbolics in July 1986. Over the next 18 months we greatly
improved the output of the development department, doubled sales,
improved customer relations (which had been pretty bad), and drove SMP
out of business. However, Symbolics' main business was sinking due
to poor strategy and management problems. So instead of giving Macsyma
more headcount to continue growing, Symbolics cut the Macsyma
In 1987-88, the Macsyma group tried to build a PC Macsyma with Gold
Hill Lisp, which proved to be too unstable. (Symbolic had killed its
Lisp compiler project for standard computers in order to avoid
competing with Lisp machine sales. They also refused to cooperate
with Sun to make Lisp available on Sun workstations for the same
reason.) When Wolfram's Mathematica was released in summer 1988,
Macsyma could not respond with a PC product because we lacked the
manpower, and because there were no good Lisp compilers on PCs.
We responded with a product in August 1989 that lacked notebooks,
and was very slow numerically. (The Lisp community at Symbolics
believed that numerical analysis was an old technology that was not
important for Lisp, so they declined to invest significant resources
to improve the inadequate speed of numerical computations of their
By 1989, it was clear that Symbolics would implode due to poor
strategy and bad management, and that they would take Macsyma
with them. I could not assemble a buy-out team to free Macsyma
from Symbolics due to lack of cooperation from M.I.T. After
trying to explain to management how to fix the main Symbolics
business (which got the new president angry), I left Symbolics.
1.3 Macsyma Inc.
In April 1992 Russell Noftsker and I founded Macsyma Inc. and
acquired the Macsyma business from Symbolics. Macsyma's market
share in symbolic math software had fallen from 70% in 1987 to
1% in 1992. While the market was growing fast, Macsyma sales
in 1991 and early 1992 were falling.
Macsyma Inc. made enormous improvements in Macsyma, culminating
in Macsyma 2.0.5 in early 1995.
o On Wester's large test of symbolic math, this product scored
10% better than Maple and 15% better than Mathematica. Instead
of being very slow, it was on average faster than Mathematica
and almost as fast as Maple.
o Macsyma 2.0.5 had a much better notebook interface than either
o All reviewers unanimously agreed that it had the best help
system in the industry (including hypertext, function templates,
and later Mathtips natural language query).
o Although Macsyma 2.0.5 was still very slow at numerics, it had
a greatly strengthened portfolio of numerical analysis and
linear algebra routines. In 1996 we added LAPACK to solve
the worst speed problems in numerical linear algebra.
By 1993, market growth had slowed and the market had standardized
on Mathematica and Maple. The competitors had development staffs
that were 4-8 times our size, so it was quite hard to gain the
lead in so many areas at once -- but we did it anyway.
Our backers were wealthy hobbyists who would not give us adequate
funds to make up for the past problems. Indeed, when we announced
Macsyma 2.0 in the fall of 1994, our primary backer cut nearly
all funding, saying he wanted to see if we could market the
product on our own, after he had paid to develop it. So we spent
virtually nothing on marketing during the year after we achieved
the greatest level of product superiority.
2. Search for Life After Macsyma Inc.
In late 1997 I had virtually nailed down an endowment of roughly
$50 million to enable Macsyma to revolutionize the interaction
of mathematics and software.
o We persuaded our potential benefactor that automated
computation was the emerging intellectual technology in
mathematics; and that, while axiom-sets-and-mappings mathematics
is extremely valuable, symbolic math software offers greater
opportunities to yield major results over the next few decades.
o Our donor told us he definitely planned to fund this plan.
o We prepared rough programs and budgets, contacted many
academics to explore interest, and we explored legal and tax
implications of various structures.
In the spring of 1998 a leading pure math department persuaded our
potential donor (an alum) that the best minds only work on pure
math. So the frontier of automated operational mathematics lost
to the mature activity of supporting bright guys to discover more
I view this result as a victory of tactical competence over
In spring 1999 I tried to persuade Florida State to adopt Macsyma
as a powerful addition to their small program in symbolic
o The project leader, Mika Seppala, told FSU that supporting
Macsyma as an open academic math system with ample resources
would eventually make FSU the world center of math software.
o The dean of science agreed. He is charged with long-term
development of the school of science. We agreed that the
best way to improve their standing is to beat the established
leaders into the emerging growth fields, and this looked like
an attractive opportunity.
o The mathematics faculty strongly rejected the proposal. They
were unwilling to see resources diverted to something other
than pure mathematics, since their resource base was already
tight due to difficulties which FSU, as a second-tier math
department, has winning research grants.
So the initiative collapsed.
2.3 Sale of Macsyma Inc.
Macsyma was acquired in 1999 by Tenedos LLC, a small holding
company. Tenedos is controlled by an ex-government agent, who
hoped to get government funds from the defense and intelligence
agencies to do something with Macsyma. As of 2003, Tenedos has
not raised funds to to anything with Macsyma, so it lies dormant.
3.1 Products and Technology
Macsyma's most damaging product problem was slowness in numerical
analysis. Whenever a customer considered Macsyma for adoption in
a major commercial or government or educational program, the
severity of this problem killed Macsyma's chances. This problem
had two causes.
o Lisp systems had slow numerical analysis. Lisp's support for
uniform garbage collection and run-time type checking (and
possibly other developer-centric features) required software
indirection that slowed arithmetic and basic array operations.
Lisp developers focused on "artificial intelligence" and they
considered numerical analysis of little consequence for Lisp.
o Macsyma built all matrices at user level from list structures
which are terribly slow for linear algebra. For those matrix
operations that were performed internally on arrays, the
conversion between lists and arrays was itself very slow.
The slowness of numerical linear algebra was too big a problem
to tackle, given our tie to Lisp and the seriousness of other
problems that needed attention after Symbolics milked the product
two times in the 1980s. So I hoped I could build a product that
was best at everything and had all the basic numerical analysis
but was slow at numerics.
3.2 Societal Support Structures
The most basic reason long-term for the decline of Macsyma was
that the academic and government mathematical and scientific
communities took virtually no interest in symbolic math software.
The academic mathematical community is trapped in an historical
lacunae of excessive abstraction that will become a backwater
within mathematics in the 21st century (or else most of mathematics
will become a backwater in the sciences).
Philanthropists who might have helped Macsyma did not, for two
o Very little philanthropy goes to the hard sciences today except
for life sciences, because these fields cannot make a compelling
case to philanthropists. I believe symbolic math software should
be the exception -- it can rejuvenate the crown intellectual
jewel of science by injecting revolutionary doses of
o The main philanthropic investor we were fortunate enough to
find eventually took the advice of the academic mathematics
community, so he funded a mathematics foundation whose programs
reflect the strategic failures mentioned above in the academic
3.3 Macsyma Team
The Macsyma team that started at M.I.T. became riven with factions
starting in the late 1970s. Not only did this precluded cooperation,
but it aligned some major institutions against one another, which
scared off investors who might have tried to get involved.
> -----Original Message-----
> From: firstname.lastname@example.org
> [mailto:email@example.com] On Behalf Of Stavros Macrakis
> Sent: Thursday, October 30, 2003 2:37 PM
> To: 'Richard Fateman'
> Cc: firstname.lastname@example.org; 'Michael Reimpell'
> Subject: RE: XML, why Mma is popular, was...Re: [Maxima]
> Lunch with George Carrette
> > The key lesson of Mathematica's success is, I think,
> > disturbing. That is, marketing and glitz was more important
> > than efficient, reliable and comprehensive mathematical
> > computation. Technical achievement was not critical since
> > most came years later.
> I find the lesson instructive, but not disturbing. It is not
> enough to have a good (or even the best) product. You must
> convince potential users to try it, and then adopt it. You
> must make the product not only be good, but appear good.
> It's probably a good idea to throw away the first draft (SMP
> was Wolfram's first attempt to improve on Macsyma,
> Mathematica his second -- was Macsyma rearchitected in the
> meantime?). You must ship on appropriate platforms. You
> bootstrap whatever success you do have into evolving both the
> product and the marketing.
> I would be fascinated to learn the full story, but I get the
> impression that Mathematica's success comes as much from
> Macsyma's marketing errors as from Mathematica's marketing triumphs.
> Maxima mailing list
> http://www.math.utexas.edu/mailman/listin> fo/maxima