B AOInts

B.2 Historical Perspective

Prior to the 1950s, the most difficult step in the systematic application of Schrödinger wave mechanics to chemistry was the calculation of the notorious two-electron integrals that measure the repulsion between electrons. Boys112 showed that this step can be made easier (although still time consuming) if Gaussian, rather than Slater, orbitals are used in the basis set. Following the landmark paper of computational chemistry111 (again due to Boys) programs were constructed that could calculate all the ERIs that arise in the treatment of a general polyatomic molecule with s and p orbitals. However, the programs were painfully slow and could only be applied to the smallest of molecular systems.

In 1969, Pople constructed a breakthrough ERI algorithm, a hundred time faster than its predecessors. The algorithm remains the fastest available for its associated integral classes and is now referred to as the Pople-Hehre axis-switch method.740

Over the two decades following Pople’s initial development, an enormous amount of research effort into the construction of ERIs was documented, which built on Pople’s original success. Essentially, the advances of the newer algorithms could be identified as either better coping with angular momentum (L) or, contraction (K); each new method increasing the speed and application of quantum mechanics to solving real chemical problems.

By 1990, another barrier had been reached. The contemporary programs had become sophisticated and both academia and industry had begun to recognize and use the power of ab initio quantum chemistry, but the software was struggling with “dusty deck syndrome” and it had become increasingly difficult for it to keep up with the rapid advances in hardware development. Vector processors, parallel architectures and the advent of the graphical user interface were all demanding radically different approaches to programming and it had become clear that a fresh start, with a clean slate, was both inevitable and desirable. Furthermore, the integral bottleneck had re-emerged in a new guise and the standard programs were now hitting the N2 wall. Irrespective of the speed at which ERIs could be computed, the unforgiving fact remained that the number of ERIs required scaled quadratically with the size of the system.

The Q-Chem project was established to tackle this problem and to seek new methods that circumvent the N2 wall. Fundamentally new approaches to integral theory were sought and the ongoing advances that have resulted980, 25, 237, 166, 817 have now placed Q-Chem firmly at the vanguard of the field. It should be emphasized, however, that the 𝒪(N) methods that we have developed still require short-range ERIs to treat interactions between nearby electrons, thus the importance of contemporary ERI code remains.

The chronological development and evolution of integral methods can be summarized by considering a time line showing the years in which important new algorithms were first introduced. These are best discussed in terms of the type of ERI or matrix elements that the algorithm can compute efficiently.

1950 Boys 112 ERIs with low L and low K
1969 Pople 740 ERIs with low L and high K
1976 Dupuis 243 Integrals with any L and low K
1978 McMurchie 627 Integrals with any L and low K
1982 Almlöf 32 Introduction of the direct SCF approach
1986 Obara 672 Integrals with any L and low K
1988 Head-Gordon 355 Integrals with any L and low K
1991 Gill 297, 302 Integrals with any L and any K
1994 White 980 J matrix in linear work
1996 Schwegler 817, 818 HF exchange matrix in linear work
1997 Challacombe 166 Fock matrix in linear work