Reflections on John Pople's Career and Legacy

by Michael J. Frisch
17 March 2004

During his more than 50 years in chemistry, John really had at least f
our careers, working in:
Statistical Mechanics
NMR
Semi-Empirical Theory
Ab Initio Electronic Structure Theory.

Moreover, his accomplishments in each area would be remarkable had tha
t been his only area of work.
From Statistical Mechanics to NMR to Semi-Empirical Methods
Although John wrote a few early papers on computational electronic str
ucture theory while a graduate student with Sir John Lennard-Jones [1]
, his primary interest for his first decade of research was in Statist
ical Mechanics. The model for liquid water he published in the 1950's
[2] remained the standard for many years.
Next, John became interested in the then-emerging field of NMR, publis
hing important papers on the underlying theory [3] and also coauthorin
g the then-standard textbook on the subject [4]. As this work progress
ed, he developed an interest in computing properties such as chemical
shifts, which lead him to electronic structure. Since first principles
(non-empirical) methods appeared to be far too expensive computationa
lly to apply to typical problems in organic chemistry, John used and d
eveloped semi-empirical models. Simultaneously with Pariser and Parr,
John developed what became known as the PPP model for p -excitations [
5], one of the earliest successful semi-empirical models. In spite of
its simplicity, this model still has its uses 40 years later.
Eventually, John's work with semi-empirical models drew him away from
NMR altogether. Along with various students, he developed the widely u
sed CNDO [6] and INDO [7] models. In the course of this work, John art
iculated and clarified the approximations used both in this generation
of models [8] and by other researchers such as Michael Dewer in later
models such as MINDO/3 [9], MNDO, and AM1. From the beginning of his
semi-empirical electronic structure work, John had the goal of creatin
g computer programs which would be useful to chemists who were not exp
erts in the theory. His CNDO/INDO program was one of the most popular
distributed through the Quantum Chemistry Program Exchange.
Electronic Structure Theory
After several years of developing successively more accurate and more
computationally costly semi-empirical models, John realized that with
improvements to the algorithms, it would be possible to make non-empir
ical (by then called “ab initio”) calculations fast enough to apply
to significant problems [10]. This type of theory remained his focus f
or the final three decades of his work. His work on the STO-3G basis s
et remains among his most cited papers [11].
After he and his student Warren Hehre had developed a new algorithm wh
ich made Hartree-Fock calculations much faster than had previously bee
n thought possible, the resulting program, Gaussian 70 [12], was made
available through QCPE. Earlier programs by other theorists (e.g., Pol
yAtom) had been distributed to and used by many theoretical research g
roups, but because of both its speed and ease of use, Gaussian 70 beca
me the first ab initio program used by significant numbers of non-theo
rists.
During the 1970's, John and his group worked on more sophisticated ab
initio methods, including larger basis sets (6-31G, 6-31G* [13], etc.)
and going beyond Hartree-Fock to include electron correlation, examin
ing several rival methods which had been advocated by different resear
chers, including Configuration Interaction, Perturbation Theory and Co
upled Cluster [14].
Through the 1970's and 1980's John was one of the leaders (along with
Bartlett and Schaefer) in the development of models and algorithms for
SCF, CI [15], perturbation theory [16], and coupled cluster methods.
One of his most notable accomplishments was a 1979 paper with Schlegel
, Raghavachari, and Binkley which presented both the first practical a
lgorithm for analytic Hartree-Fock second derivatives and the first gr
adients for the MP2 correlated method [17]. Work in John's group conti
nued in these areas, as well as on improved algorithms for integral ev
aluation, Hartree-Fock calculations, resulting in the HGP [18] and PRI
SM [19] algorithms, and on direct and semi-direct algorithms for large
MP2 calculations [20].
In the 1990's, John saw that Becke had applied the Model Chemistry app
roach (see below) to various density functional models and demonstrate
d that some of these functionals had sufficient accuracy to be useful
for chemical problems, and began to work in this area as well [21]. Wo
rk in recent years has focused on the development of the high accuracy
Gaussian-1 theory [22] and its follow-ons.
Model Chemistries: John Pople's Legacy
In approaching each of these methods in turn, John was guided by his p
rinciple of Model Chemistries , an original concept for which he was s
olely responsible and articulated in a hard-to-find seminal paper [23]
. This approach, in which one carefully calibrates the difference betw
een the chemistry predicted by a particular model and that observed in
the real world, and then uses the same model for studies of new syste
ms in which the accuracy--and error--of the model is known from the pr
evious calibrations, represented a significant departure from the appr
oach taken in earlier theoretical work.
Traditionally, theorists tried to do the best calculation they could o
n a particular problem, using different models, basis sets, and so on
for each study. As a result, the accuracy of a new calculation was dif
ficult to evaluate even for experts (and impossible for non-expert to
even estimate). The consequence so this were two-fold:
It effectively prevented the application of electronic structure calcu
lations to a broad range of problems.
It made it extremely difficult for non-experts to use electronic struc
ture calculations as part of their research.
By articulating the principle of model chemistries and then carefully
calibrating particular models, John made it possible for people to app
ly computational models with confidence because they knew to what exte
nt to trust the results. This fundamental principle has been essential
to the success of almost all methods in electronic structure theory,
not just those originated or favored by John, and has also been essent
ial to the widespread use of electronic structure computations, regard
less of which software package is involved. This is undoubtedly John P
ople's most significant contribution, and will still influence the dev
elopment of the field long after the particular models he developed ha
ve become obsolete.
I started as a graduate student of John in 1979, in what turned out to
be the middle of his career. Actually, I was one of a succession of g
raduate students each of whom was widely assumed to be John's last stu
dent, but fortunately this turned out to be far from the case. While I
learned about many technical aspects of theory from John, I think tha
t the most valuable wisdom he passed on to me and his other students a
nd post-docs consisted of the principles he applied to his research an
d the approach he advocated:
Theorists should compute what is measured not just what is easy to cal
culate
Theorists should study systems people care about, not just what is eas
y or inexpensive to study.
Models should be calibrated carefully and the results presented with s
crupulous honesty about their weaknesses as well as their strengths.
One should recognize the strengths as well as the weaknesses of other
people's models and learn from them.
If a model is worth implementing in software, it should be implemented
in a way which is both efficient and easy to use. There is no point i
n creating models which are not useful to other chemists.
These ideas seem as sound to me today as they did when I first learned
them from John more than twenty years ago. The goal of making theory
a useful tool for all chemists has clearly been adopted by many of his
students. As John was very fond of pointing out, his former group mem
bers have started a total of five software companies. I would add the
note that since all five companies are still in operation after a deca
de or more, that the people involved have also learned some of John's
other lessons about how to do theory in a way that matters to the fiel
d as a whole.
As is often the case with great scientists, John demonstrated a keen a
bility for putting aside the many non-essential details of a complicat
ed problem, and identifying and focusing on the critical issues. I'm s
ure that all his students try to follow his example in this, but that
none of us would claim to have his talent at it. His passing is a loss
to everyone in our field.
--
FROM 10.22.11.62