J . Phys. Chem. 1990,94, 5439-5444 not by the Pople group which adhered to the philosophy of constructing a well-defined model which was then to be understood for its strengths and limitations. Thus Del Bene and Jaffe quickly moved to a parametrization CNDO/S'O which was particularly well suited to the calculation of electronic excitation energies and oscillator strengths for conjugated systems including the effects of polarity in heteroatomic molecules. Similarly, Dewar" and =workers immediately put forth the first of what were to be many generations of M I N D 0 methods which sought develop a theory based upon INDO which would allow one to reliably to predict heats of formation. This, of course, has betn extremely successful and is widely used today. It is an available option in computer programs like GAUSSIAN 86. John, however, moved on to ab initio methods, and his group produced GAUSSIAN 70, which was to quantum chemistry what the IBM PC was to personal computers-it organized and formed (IO) Del Bene, J.; JafE, H. H. J . Chem. Phys. 1968.48, 1807; 1968,49, 1221. (11) Baird, N. C.; Dewar, M. J.
S.J . Chem. Phys. 1968,50, 1262.
5439
the field. That transition, however, is the subject of the next talk and it is time for me to finish. John Pople has been a leading figure in theoretical chemistry from the first classic papers published by the Lennard-Jones group at Cambridge in the early 1950s, through the NMR years, to his seminal effect on computational quantum chemistry. During the time that I was with him, John received an award-I cannot remember what it was and John cannot remember either. At any rate, when I congratulated him on it, it was obvious that the fact that he was just turning forty was bothering him, because he answered, "When this sort of thing starts happening, you know you're getting old". (Let me tell you, when somebody asks you to give an historical talk on your thesis, you know you're getting old.) In John's case, we know that he had nothing to fear-he was to be a fountain of new and imaginative science for the next twenty-five years and today we honor the forty years of his career so far. As I said in the beginning, I feel fortunate to have been one of those to benefit directly from his influence and I want to thank Fritz and Nick for giving me the opportunity to dwell on an important piece of work in John's career and an important period of my life.
John A. Pople: Early ab Initio Days Leo Radom Research School of Chemistry, Australian National University, Canberra, A.C.T.2601, Australia (Received: November 9, 1989)
An account is presented of research activities in John Pople's group at Carnegie-MellonUniversity from 1968 through 1972. The development of GAUSSIAN 70 during this period meant that ab initio calculations could be carried out straightforwardly and rapidly. This opened the way for applications of ab initio theory on a scale that would previously not have been feasible and laid the foundations for the widespread use of ab initio calculations by the chemistry community.
When I received the invitation to speak at the symposium Forty Years of Quantum Chemistry: An International Conference in Honor of Professor J. A. Pople, I had mixed emotions. On the negative side, this was the first occasion on which I had been asked to give a historical talk, and it made me realize that time does not stand still and that age must be catching up with me. On the positive side, I was delighted to be able to contribute to a symposium that deservedly honors my mentor, colleague, and friend, John Pople. So thank you to the organizers, Fritz Schaefer and Nicholas Handy, for having provided me with the opportunity to participate, and congratulations to them also for having brought together a t this conference such an impressive array of talent. My brief in this presentation is to discuss "the early ab initio days". I will be examining a period of roughly 5 years from 1968 through 1972. I should stress at the outset, so as not to offend anyone, that I am not attempting to review the advances in quantum chemistry that took place outside of Pittsburgh. This would be an impossible task in the time available. Rather, I will be focusing entirely on events in the Pople group a t CarnegieMellon University during this period. I spent 3l/* of those years beginning in early 1969 as a postdoc in the group, and a very pleasant and stimulating time it was. But the story begins before I arrived in Pittsburgh, and we should probably take it up in early 1968. At that time, the series of all-electron semiempirical S C F schemes were already firmly established' and were offering an economical alternative to the a b initio calculations that were being carried out in various laboratories, mainly in the U.S.but also elsewhere in the world. A (1) For a review of the semiempirical procedures. see: Pople, J. A.; Beveridgc. D. L. Approximate Molecular Orbital Theory; McGraw Hill: New York, 1970.
ballpark figure of about 1:lOOO would have described the relative computing costs of say CNDO and minimal basis set ab initio calculations at the time. John Pople saw a natural progression from the semiempirical to the ab initio approach. As a first step in attempting to bridge the gap, he introduced, together with Marshall Newton and Neil Ostlund, the approximate ab initio PDDO (or projection of diatomic-differential-overlap) schemee2 This involved approximating some of the more problematical integrals using a generalization of the Mulliken approximation. The PDDO method was slower than CNDO (by a factor of about 10)but produced useful savings in computation time compared with the fully ab initio approaches that were available at the time. However, important developments in the fully ab initio procedures soon took up the running. They had their origins in the work of Boys, who, some 20 years earlier,3had suggested that the integral bottleneck in ab initio calculations, the evaluation of four-center two-electron integrals of the type (4&1J#J~&)involving basis functions &, could be overcome by using Gaussian basis functions (exp(-d)) instead of exponential basis functions (or Slater-type atomic orbitals, STOs). The Gaussian functions had the crucial property that the product of Gaussian functions at two centers A and B was itself a Gaussian function at a third center C. The disadvantage of using Gaussian basis sets was that several Gaussian functions were needed to reproduce the results of a single (2) (a) Newton, M. D.; Ostlund, N. S.;Pople, J. A. J . Chem. Phys. 1968, 49, 5192. (b) Newton, M. D. J . Chem. Phys. 1969, 51, 3917. (c) Newton, M. D.; Lathan, W. A.; Hehre, W. J.; Pople, J. A. J . Chem. Phys. 1969, 51, 3921. (3) Boys, S. F. Proc. R.Soc. London 1950, A200, 542.
0022-3654/90/2094-5439$02.50/00 1990 American Chemical Society
5440 The Journal of Physical Chemistry, Vol. 94, No. 14, 1990 STO. Thus, the individual or primitive Gaussian functions, g k gk
= x exp(-ak$)
(1)
were grouped together as linear combinations to form the basis functions q$:
6fi = Cdk&k
(2)
The 4, were known as contracted Gaussian functions. However, the much greater efficiency of integral evaluation using the Gaussian basis functions more than made up for their larger number. There were several ab initio program packages in use at the time that did indeed utilize Gaussian basis functions (e.g., POLYATOM, IBMOL), but they were, as I have already mentioned, much slower than the semiempirical procedures and their a p plication was computationally expensive. In a visit to the University of Florida in May 1968, Pople shut himself away for a few days and emerged with algorithms that would lead to improvements by a factor of about lo0 in the integral evaluation times.’ Two features were exploited to achieve this computational efficiency. The first was to group together in shells basis functions that share common information. As part of this strategy, the s and p functions on an atom were constrained to have the same exponent, allowing as a consequence s, px, pp and pz to be grouped in an sp shell. Integrals involving such related basis functions were then evaluated together, e.g., 44 = 256 integrals for the integral (ABICD) involving four sp shells. The second strategy arose from the recognition that if each basis function 4Nis the sum of K primitive Gaussian functions gk, there are K4 primitive integrals (gg&g,) for each two-electron integral (+N&l#Aq5u). One should therefore reduce as much as possible the computations that have to be performed K4 times. This was achieved by axis transformations within inner loops. Pople returned to Pittsburgh where, with the assistance of his very capable and hard-working graduate student Warren Hehre, about whom I will say more later, these ideas were incorporated into an ab initio program that eventually became GAUSSIAN 70.’ Hehre had a working version of POLYATOM for comparison and, in test calculations, found that GAUSSIAN 70 produced an improvement of about 2 orders of magnitude in the time required to calculate the two-electron integrals. In order to make full use of the special features of this new ab initio program, special basis sets needed to be developed. Bob Stewart, a faculty member at CamegieMellon University, played an important role in this process. Stewart had been engaged in constructing basis sets for the purpose of evaluation of generalized scattering factors for X-ray diffraction studiesa6 His approach involved using a least-squares procedure to fit linear combinations of Gaussian functions to STOs. So he was just the right person to call on for help in the development of the new basis sets for GAUSSIAN 70. In this way STO-3G, probably the most widely known basis set in the history of quantum chemistry, was born.’ STO-3G was a minimal basis set in which the individual basis functions comprised a least-squares fit of three Gaussian functions to the appropriate STO. STO-3G had the special feature, which I noted above, that common Gaussian exponents were shared between the 2s and 2p orbitals for computationalefficiency. The least-squares fits were obtained for STOs with exponents of unity and then rescaled as appropriate. Several important results were found in this early study. First, the results for Gaussian expansions of inmasing size (Le., STO-NG with N = 2, 3,4, 5, and 6) showed rapid convergence toward the STO limit. On this basis, STO-3G was chosen as a reasonable compromise between cost and accuracy. Second, on the basis of optimization of exponents for a number (4) Hehre, W. J.; Pople, J. A. J . Compur. Phys. 1978, 27, 161. ( 5 ) Hehre, W. J.; Lothan, W. A.; Newton, M. D.; Ditchfield, R.; Pople, J. A. Quontum Chemisrry Program Exchange: Bloomington, IN, 1973; program no. 236. (6) Stewart, R. F. J. Ch” Phys. 1969, 50, 2485. (7) (a) Hehre. W. J.; Stewart, R. F.; Pople, J. A. J . Chem. Phys. 1%9, 51, 2657. (b) Hehre. W. J.; Ditchfield, R.; Stewart, R. F.; Pople, J. A. J . Chem. Phys. 1970, 52, 2169.
Radom of small molecules, a set of standurd exponents was selected leading to a standard STO-3G basis set. STO-3G calculations for small organic molecules were about 10-20 times slower than CNDO. Choice of basis set is one of the most important aspects of ab initio molecular orbital calculations, and considerable effort was expended on developing suitable basis sets in the Pople group, the leading players being Warren Hehre and Bob Ditchfield. STO-3G was, as I mentioned, a minimal basis set (i.e., comprised, with some minor adjustments, of exactly that number of functions required to accommodate all of the electrons of the atom while maintaining overall spherical symmetry) obtained through a least-squares procedure. A series of alternative minimal basis sets were constructed by energy minimization rather than through least-squares fitting8 These had names such as LEMAO-3G meaning least-energy minimal atomic orbitals. (I suggested that they simply be called least-energy orbitals, with the simpler acronym LEO-3G, but this suggestion was not greeted with enthusiasm!) These basis sets performed less well than the STO-NG sets and so were not widely used. One of the shortcomings of the minimal basis sets was that the sizes of the orbitals were predetermined by the choice of exponents. This was an unsatisfactory state of affairs because clearly the ideal size of an orbital should depend on the environment. In addition, prescribing fixed, equal exponents to each of the p-orbital components meant that an isotropic distribution of orbitals was being imposed. Differences between, for example, p,. and p, orbitals could not be accommodated under this scheme. These problems could be overcome, however, by using two functions instead of one to describe the valence orbitals, i.e., by splitting the valence basis functions into two parts. The sizes of the orbitals could then adjust variationally to the molecular environment (1); anisotropy
contracted
diffuse
1
of the electron distribution would also be possible since, for example, the p,. and p,, orbitals could become unequal in size. The best known of the early split-valence basis sets was 4-31G, in which four Gaussian functions were used for each orbital but with the valence orbitals split into two parts containing three and one Gaussian function respecti~ely.~The 6-31G basis’* was a slightly larger split-valence basis but was not often used in its own right; its main use was as the starting point for even larger basis sets. The final improvement to the basis sets that was accomplished during this period was the addition of so-called polarization functions. The aim was to allow nonsymmetrical electron distributions about atoms, i.e., polarization, and this was achieved by the addition of p functions on hydrogen atoms and d functions on first-row atoms to the 6-31G split-valencebasis set (e.g., 2).
P. C. Hariharan wrote the program to calculate the integrals involving d functions, and this led to the creation of the well-known split-valenceplus polarization 6-31G* and 6-31G** basis sets.“ Basis sets of this type were found to be very important in describing cyclic or bridged systems. I should say something at this stage about the computing facilities that were available to us at the time. We used a CDC-1604, one of whose claims to fame was that it was the first computer (8) (a) Ditchfield, R.; Hehre, W. J.; Pople. J. A. J . Chem. Phys. 1970.52, 5001. (b) Hehre, W. J.; Ditchfield, R.; Pople, J. A. J. Chem. Phys. 1970,53. 932.
(9) Ditchfield, R.; Hehre, W. J.; Pople, J. A. J . Chem. Phys. 1971.54724. (IO) Hehre, W. J.; Ditchfield, R.; Pople, J. A. J . Chem. Phys. 1971, 56, 2257. ( 1 1) Hariharan, P. C.; Pople, J. A. Theor. Chim. Acra 1973, 28, 213.
John A. Pople: Early ab Initio Days to be designed by Seymour Cray. That is where the good news ends. The 1604 had a delivered speed of about 0.03 Mflops. To put this in perspecitve, it is less than half that of a Macintosh 11. The amount of memory available was 192 kbytes, in this case less than 20% of that of a simple Macintosh Plus! The storage medium was magnetic tape, which was both messier and considerably less reliable than the disk storage of today. The program was punched out on cards, using a machine that would be hard to find these days called a card punch. There were trays upon trays of cards, so loading a new version of the program was always a tiresome procedure. One distinctive feature of the CDC- 1604, which Marshall Newton and Janet Del Bene recently reminded me of, was an audio accessory that converted the digital processing into distinctive audio signals. You could actually hear the ab initio calculations being crunched out. Marshall Newton speaks of “the cascading cadences produced by the calculation of the incomplete gamma function table a sure-fire indication that the Gaussian program was once again being pressed into service”. Why was it that GAUSSIAN 70 was so important? In my view, the key point is that it made possible an approach to ab initio calculations very different from that which was prevalent at the time. Let me elaborate. There were in the late 1960s some very active groups carrying out a wide variety of interesting applications. However, because of computational expense, most of the papers at the time were concerned with calculations on one or at most a small number of molecules. This changed with the advent of GAUSSIAN 70 where the speed and, very importantly, the userfriendly design enabled applications of ab initio theory, even with quite modest computing resources, to be performed on a scale that was previously not feasible. It was now possible to carry out ab initio calculations on large numbers of molecules at a uniform level of theory. This enabled the development of what Pople calls theoretical-model chemistry.12 A level of theory is first clearly defined and then applied uniformly to a wide variety of molecules up to a maximum size determined by available computational resources. If the model performs satisfactorily in systematic comparisons with available experimental data, it then acquires some predictive value in situations where experimental data are not available. The HartreeFock/STO-3G model was the first of a (continuing) sequence of such ab initio theoretical models. Such an approach has the attraction that one has some idea of the reliability of the results. In addition, the ability to carry out large numbers of calculations at a uniform level means that comparative data can be obtained. Chemistry is largely a comparative science and so such data are vital in making generalizations. I will present relevant examples shortly. Let me pause at this stage and describe the players on the Pople team during the 1968-1 972 period. I already mentioned Marshall Newton’s role with PDDO in the transition from semiempirical to ab initio approaches. Newton was also involved in an early study at the STO-3G level of the geometries and quadratic force constants of some 32 polyatomic molecules containing up to 4 heavy atoms.13 The average deviation from experiment for bond lengths was 0.035 A and that for bond angles 1.7O. Thus, STG3G was shown to provide an economical approach to predicting molecular geometries, generally to reasonable accuracy. It is interesting that some of the worst performing molecules for STO-3G, notably those containing bonds between highly electronegative elements such as F202,continue to prove problematical to theoreticians even today. Newton left Pittsburgh at the end of 1969 to take up an appointment at Brookhaven National Laboratories. Next, we have Warren Hehre. Warren was one of the first members of the group I met when I arrived in Pittsburgh, and I remember asking him at the time what hours people tended to work. You see, I had been told Americans worked rather longer
...
(12) Poplc. J. A. Ace. Chem. Res. 1970, 3, 217. (13) Newton, M. D.; Lathan, W. A.; Hehre, W. J.; Pople, J. A. J. Chem. Phys. 1970, 52,4064.
The Journal of Physical Chemistry. Vol. 94, No. 14, 1990 5441 hours than we did back in Australia and I did not want to be out of step. I was concerned when Warren said that he worked 7 days a week but was relieved when he said that he only worked from 9:00 till 3:00. It only took me a couple of days, however, to discover what 9:00-3:00 really meant! Warren Hehre was the linchpin of the group. He was involved in most of what was going on and certinly knew everything that was going on. He could be described as the quarterback on the team, although my scant knowledge of American football suggests that he was not of the conventional shape. He had a major involvement in program development, in the construction of the various basis sets, and in many of the applications that I shall mention shortly. Hehre completed his thesis in 1971 and spent a postdoctoralperiod with Lionel Salem in Orsay before joining the faculty at the University of California at Irvine. Bill Lathan was another graduate student in the group. His task was to optimize the geometry of every molecule in existenceor so it seemed. In fact, he restricted himself largely to optimizing the structures of all molecules with one or two of the first-row elements carbon through fluorine. Some giant papers resulted. One was a systematic study, using STO-3G and 4-31G, of AH,, molecules and cations for first-row atoms A = C, N, 0, and F.I4 This included calculations with full geometry optimization of some 70 systems, among which were some very topical species such as CH4*+and CH5+. Among others, there was a systematic study, using STO-3G, 4-31G, and 6-31G*, of the 30 possible three-membered ring molecules containing C, H, N, and 0.15This included calculation of geometries, orbital energies (for comparison with photoelectron spectra), and strain and resonance energies. On an even grander scale, there was a systematic study of HJBH,, systems.16 This involved neutral and positively charged systems containing two of the first-row atoms C, N, 0, and F and any number of hydrogen atoms, a total of approximately 300 species for which geometries were optimized! It is not feasible to discuss individual results here, but it goes without saying that this paper produced a massive amount of data, some of which could be compared with existing experimental information but much of which was predictive for systems that had not yet been experimentally characterized. Many of the predictions have been useful in subsequent experimental studies. Lathan left Pittsburgh for a postdoctoral position with Keiji Morokuma, then at the University of Rochester, before joining a Department of Defense establishment in Nebraska. John Lisle was also a contributor to this last work.I6 I should note that one of John Lisle’s main nonchemical claims to fame was that he had an even heartier appetite than Warren Hehre. There was one memorable group outing to a Howard Johnson’s restaurant on a Wednesday night where they offered all the fish you could eat for $1.29. When Hehre and Lisle decided to have a competition to see who could eat the most, it was like the irresistible force against the immovable object! I think they were up to the twentieth piece of fish before a truce was called. John Lisle left Pittsburgh for the University of Connecticut in 1971. Don Miller was another graduate student in the group. His work, which was largely carried out in collaboration with Bob Ditchfield, was concerned with calculating first- and second-order properties of small molecules.17 After completing his thesis, Miller joined Gary Maciel’s group at Colorado State University. Janet Del Bene came to the group on a postdoctoral fellowship in 1969, and the main emphasis of her work in Pittsburgh was in the area of hydrogen-bonded polymers of small molecules such as water and hydrogen fluoride.l8 Del Bene and pople found a (14) Lathan, W. A,; Hehre, W. J.; Curtis. L. A,; Pople, J. A. J. Am. Chem. Soc. 1971, 93,6317. (15) Lathan. W . A.: Radom. L.; Hariharan. P. C.; Hehre, W. J.; Poole, J. A.Fortschr. Chem. Forsch. 1973, 40, 1. (16) Lathan, W. A.; Curtis, L. A.; Hehre, W. J.; Lisle, J. B.; Pople, J. A. Prog. Phys. Org. Chem. 1974,11, 115. (17) (a) Ditchfield. R.; Miller, D. P.; Pople, J. A. J . Chem. Phys. 1970, 53, 613. (b) Ditchfield. R.; Miller, D. P.; Pople, J. A. J . Chcm. Phys. 1971, 54, 4186.
5442 The Journal of Physical Chemistry, Vol. 94, No. 14, 1990
cooperative effect in the hydrogen bonding in both the H 2 0 and H F polymers which led to a nonadditivity of hydrogen-bond energies. In addition, for all polymers other than the dimers, they predicted cyclic structures to be more stable than corresponding open-chain structures and of sufficient stability that they should exist in gaseous H 2 0or HF along with the monomers and dimers. The Del Bene and Pople studies on polymers of water were being carried out at just the right time. This was 1969 and the by the Russian chemist Deryagin of an anomalous form of water, or polywater as it was known at that time, was receiving great publicity. It must have been very tempting indeed to jump on the bandwagon! However, Del Bene and Pople found no evidence from their calculations for specially stable, symmetric forms of water polymers, and therefore, unlike some others, they resisted the temptation. The analogy with the recent flurry of activity on cold fusion is quite apposite. Janet Del Bene left Pittsburgh to take up a faculty appointment at Youngstown State University in 1970. Larry Curtiss started his graduate studies in the fall of 1969. He was a major contributor to the AH,, and H,,,ABH,, paper^.'^.'^ His main unique contribution, however, while in Pittsburgh involved some pioneering studies of the vibrational spectra of hydrogen-bonded complexes.” These included studies of the vibrational force fields of the water dimer, the hydrogen fluoride dimer, and the HCN-HF dimer. Curtiss moved to Battelle in 1973 before joining Argonne National Laboratory in 1976. P. C. Hariharan was the group expert on d functions. His program was invaluable in a number of applications, including some that I shall mention in a moment, in which the use of basis sets that included polarization functions was essential if reasonable results were to be obtained.2’ Hariharan proceeded to a postdoctoral position with Shi Shavitt at The Ohio State University in 1973, then worked under Werner Kutzelnigg in Bochum, and finally joined Joyce Kaufmann’s group at Johns Hopkins University. Bob Ditchfield joined the Pople group after completing a Ph.D. with John Murre11 at the University of Sussex. He had an outstanding general knowledge of theory and made contributions to a wide variety of projects. I have already talked about Ditchfield‘s involvement in basis set development with Warren Hehreg-’I and his work on molecular properties with Don Miller.” There was also work on excited states with Janet Del Bene.= Finally, there was work on bond separation reaction^.^^ Ditchfield left Pittsburgh in 1971 to join Larry Snyder’s group at Bell Laboratories before moving to a faculty position at Dartmouth College. Bond separation reactions were defined23as formal reactions in which a molecule is separated into its simplest parents containing the same component bonds. For example, the bond separation reaction for propene is CH3--CH=CH* + CH4 CH3-CH3 + CH2=CH2 (3) It can be seen that the C-C single and double bonds of propene are “separated” in the products, ethane and ethylene. The energies of bond separation reactions, or bond separation energies, provide a systematic means of measuring the interactions between bonds. For example, in the case of propene, the small positive bond separation energy (4.0 kcal mol-’ with STO-3G or 5.0 kcal mol-’ +
(18) (a) Del Bene, J.; Pople, J. A. Chem. Phys. Leu. 1969, I ,426. (b) Del Bene, J.; Pople, J. A. J. Chem. Phys. 1970,52,4858. (c) Dd Bene, J.; Popie, J. A. 1.Chcm. Phys. 1971,55, 2296. (d) Del Bene, J.; Pople, J. A. J. Chem. Phys. 1973,58,4858. (19) Deryagin, B. V.; Churayev, N. V. Prirdu (Moskow) 1968, I , 16. (20) (a) Curtis, L. A.; Pople, J. A. J . Mol. Spctrosc. 1973,48,413. (b) Curtiss, L. A.; Pople. J. A. J. Mol. Spcrrosc. t975,55, 1 . (c) Curtis, L. A.; Pople, J. A. J . Mol. Spectrosc. 1976, 61, 1. (21) (a) Hariharan, P. C.; Lathan, W. A.; Pople, J. A. Chem. Phys. Lcrr. 1972, 14, 385. (b) Hariharan, P. C.; Pople, J. A. Chem. Phys. Lerr. 1972, 16, 217. (c) Hariharan, P. C.; Pople, J. A. Mol. Phys. 1974, 27, 209. (22) (a) r)el Bene, J.; Ditchfield, R.; Pople, J. A. J. Chcm. Phys. 1971,55, 2236. (b) Ditchfield, R.; Del Bene, J.; PopIe, J. A. J. Am. Chem. Soc. 1972, 94, 703. (c) Ditchfield, R.; Del Bene, J.; Pople, J. A. J. Am. Chem. Soc. 1972, 94,4806. (23) (a) Ditchfield, R.; Hehre, W. J.; Pople. J. A.; Radom, L. Chem. Phys. Lert. 1970, 5, 13. (b) Hehre, W. J.; Ditchfield, R.; Radom, L.; Pople, J. A. 1. Am. Chem. Soc. 1970, 92, 4196.
Radom experimentally) reflects the hyperconjugative interaction of a methyl group with a double bond. The negative bond separation energy found for small-ring systems such as cyclopropane reflects, in part, the strain energy, and so on. If there were no interactions between bonds, the bond separation energies would all be zero. Bond separation reactions such as (3) are examples of formal chemical reactions in which the reactants and products have the same number of each type of bond but in which the relationship among the bonds is changed. The process of creating a word that would describe such reactions involved a visit to the University of Pittsburgh library. John Pople, Warren Hehre, Bob Ditchfield, and I went over together, consulted the Latin and Greek dictionaries, and came up finally with the word isodesmic, from the Greek is0 meaning equal and desmos meaning bond. The attractive prospect associated with the calculation of energies of isodesmic reactions was that the reactants and products were sufficiently similar that there was a reasonable possibility of substantial cancellation of errors associated with deficiencies, such as neglect of electron correlation or incompleteness of the basis set, in the theoretical procedure. Indeed, it was found that even quite small basis sets such as STO-3G and 4-31G yielded quite reasonable energies for isodesmic reactions in general and bond separation reactions in particular, as in the propene example above.23 The good description of bond separation reactions provided by simple levels of theory furnished a means of obtaining the heats of formation of molecules with three or more heavy atoms, provided that experimental heats of formation for the limited set of parent (one- and two-heavy-atom) molecules were known.24 For example, the heat of formation of propene could be derived from the theoretical energy of the bond separation reaction (3) together with the experimental heats of formation for methane, ethane, and ethylene. This was done for the complete set of 68 distinguishable isomers of acyclic molecules containing one, two, or three of the atoms C, N, 0, and F and which can be written as classical valence structures without charges or unpaired electrons.24 The mean absolute deviation from known experimental heats of formation was 4.8 kcal mol-’ for STO-3G and 3.1 kea1 mol-’ for 4-31G. The study of bond separation reactions of these H, C, N, 0, F molecules allowed the interactions of the bonds in these systems to be analyzed systematically. Interactions of CH3-, NH2-, OH-, and F- groups with a variety of saturated systems, interactions with unsaturated systems, interactions among cumulated bonds, and so on were examined in detail and provided extremely useful insights into the theory of substituent effects. The work on substituent effects was one of several Hehre, Radom, and Pople studies. Before discussing other work in which 1 was personally involved, I should perhaps say that I joined John Pople’s group as a postdoc in early 1969, having just completed a Ph.D. at the University of Sydney in physical-organic chemistry. I often think back to how extremely fortunate I was, coming with a background in structural and mechanistic organic chemistry to Pittsburgh where John and his team had just put together this marvellous new tool (GAUSSIAN 70) for studying organic molecules. Among other Hehre, Radom, and Pople studies was a systematic study of conformations,stabilities, and charge distributions in a set of 35 monosubstituted benzene^.^' By this time, the CDC-I604 had been replaced by Univac 1108, but it was still largely the speed of GAUSSIAN 70 that enabled such a large set of relatively complex molecules to be examined and useful comparative data to be obtained. The paper dealt with properties such as stabilization energies (i.e., energies of interaction of the substituents with the ring), dipole moments, charge distributions, and rotational barriers. An extension to disubstituted benzenes led to a useful collaboration with Bill Fateley, an infrared spectroscopist at Carne(24) (a) Hehre, W. J.; Radom, L.; Pople, J . A. 1. Am. Chem. Soc. 1971, 93,289. (b) Radom, L.; Hehre, W. J.; Pople, J. A. J . Chem. SOC.A 1971, 2299. (25) Hehre, W. J.; Radom, L.; Pople, J. A. J. Am. Chem. Soc. 1972,94, 1496.
John A. Pople: Early ab Initio Days
The Journal of Physical Chemistry, Vol. 94, No. 14, 1990 5443
gie-Mellon University at the time.% We examined the effect of para substituents on the barrier to rotation about the C-O bond in phenol and found good agreement between the spectroscopic and ab initio results. There were, in fact, a large number of additional studies of molecular conformation and rotational and inversion barriers that were pursued over those years, which I will mention by name only. These included a study of internal rotation in C2