EXTENDED HUCKEL THEORY AND MOLECULAR HARTREE-FOCK SCF THEORY* BY F. PETER BOER, MARSHALL D. NEWTON, AND WILLIAM N.

LIPSCOMB
DEPARTMENT OF CHEMISTRY, HARVARD UNIVERSITY

Communicated August 27, 1964

In the past several years a wide variety of molecules has been studied by means of the extended Htickel theory, which explicitly includes all overlaps between atomic orbitals and systematic guesses for the elements of the Hamiltonian matrix.1 These studies, in spite of their very approximate nature, have yielded a wealth of chemical information in terms of eigenvalues and eigenvectors: charge distributions and reactive sites, ionization potentials, rotational barriers, relative stabilities of isomers, stable molecular configurations, binding energies, etc. Unfortunately, a theoretical understanding of the extended Huckel method has lagged behind its often spectacular success. In this brief paper we address ourselves initially to understanding a small fragment of this success: ability to approximate binding energies. We also try to enumerate the serious errors involved in obtaining the eigenvalues, and come to the conclusion that in the present form of the method several large errors more or less cancel. We suggest that the extended Hfickel theory1 may be regarded as a method of simulating Hartree-Fock (HF) calculations by guessing the elements of the HF Hamiltonian matrix through the use of the Wolfsberg-Helmholz approximation. A critical examination of this method has been made possible by the availability of a diatomic wave function package.3 A sounder theoretical basis for the extended Huckel theory is obtained because nuclear repulsions N in the HF theory are shown to cancel approximately against half of the difference between molecular and atomic core energies, not, as expected,4 against electron-electron repulsions which are counted twice when eigenvalues are summed over electrons. This new observation is fortunate since the electron-electron term of the HF method is often a few times larger than the nuclear term.5' Define Eim as the core (kinetic plus electron-nuclear attraction) energy of an electron in the ith molecular orbital, 2(2Jij - K1j) as the electron-electron interij action energy, A as the energy of atomization (negative for stable molecules), Eim as the eigenvalues of the one-electron HF Hamiltonian, and the new quantity A as (Eim - E a) + N (where a refers to atoms). Now the HF closed-shell molecular energy is E = 2EEim + E(2Jj - Kj) + N,
t

iJ

and

,im = Eim + (2Jj - K2j). We neglect relativistic and molecular extra correlations (1\IEC) energy, and equate E for the molecule to E for the constituent atoms plus atomization energy, as follows: E = 2]Eim + tee-m + N = L:Eia + h~a + A)(1)
890

VOL. 52, 1964

CHEMISTRY: BOER, NEWTON, AND LIPSCOMB

891

so that A becomes (2) Trim = N - (ZEa - EEim), where the sums are taken over orbitals. We thus see that a necessary condition for approximating binding energies A by means of the extended Hickel theory (which does not yield a value of the total molecular energy, but only the elm) is that A be small compared to other energy terms in (2). The other conditions are, of course, that the extended-Huckel Ejm are close to the SCF values, and that the MEC contributes relatively little to A. Values of A obtained from all existing LCAO-SCF calculations are small compared with other molecular energy terms, although they depend somewhat on the choice of basis sets. A few ionic diatomics have values of A which are about ten per cent of nuclear repulsions;6a the more typical values are -0.16 au for CH4,5" 0.29 au for C2H6,5d 0.30 au for C2H4,5' and 0.58 for C2H2,5f which are -1.2, 0.7, 0.9, and 2.3 per cent, respectively, of N. Nevertheless, A is usually an appreciable fraction of A; hence, binding energies in terms of eim and eta could only be expected to have the right order of magnitude, but might be systematic within homologous series. If A were indeed zero, we would then find the formula for binding energies in the Huckel method used previously,' except for a factor of two [sums are over orbitals, not electrons in eq. (2) ], and except for the use here of the qa values6 instead of the valence state ionization potentials (VSIP). Thus we believe that the general smallness of A makes less surprising the otherwise unexpected result that the above formula' has predicted for boron hydrides"a and hydrocarbons" binding energies which are systematically larger than the experimental values by a factor of 1.8, but
A =
-

ZEia + A

otherwise internally consistent. This comparison of Hickel and SCF theories requires good values for the HF eigenvalues, eimn. We have computed the SCF Hamiltonian matrices for the diatomics BH (at 1.21 A) and BB (at 1.77 A) in preparation for a later study of matrix elements for boron hydrides. In Table 1 we compare the E," with those obTABLE 1
HF
HF core

Extended Huckel

Simulated HF

BH e, a 62 ff 83 a Z ei a BB el as E au ES ag ( 4f=u C5 7ru 2 el
*

-7.751 -0.664 -0.355 -8.770 -7.735 -7.735 -0.671 -0.384 -0.331 -16.856

(-7.678)* -0.883
-0.402 -8.963

(-7.678) (-7.678)
-0.937 -0.728 -0.297 -17.318

(-7.678) -0.667 -0.384 -8.739 (-7.678) (-7.678) -0.682

-0.454 -0.364 -16.856

-7.747 -0.789 -0.382 -8.918 -7.738 -7.737 -0.800 -0.330 -0.223 -16.828

Numbers in parentheses are Clementi's atomic 18 values, ref. 6.

tained for a HF calculation based on valence electrons only (HF core). Results are also listed for ejm as given by the extended Hickel method in which diagonal matrix elements aj are VSIP's (inner shell electrons assimilated in the effective nuclear core), and off-diagonal elements ,6, are given by the WH approximation" 7 (K = 1.75). Finally, we present results for a simulated HF Hamiltonian matrix constructed by (1) choosing diagonal elements a, from the HF calculation, (2) calculating KUj for each off-diagonal ,l of the HF matrix according to the equation

892

CHEMISTRY: BOER, NEWTON, AND LIPSCOMB

PROC. N. A. S.

= KjjSj1[(aj + a1)/2], (3) averaging Kij weighted by flij to obtain Kave, and (4) generating the off-diagonal elements from the WH equation and the overlap integrals. Thus, we attempt to reproduce the HF results as closely as possible within the framework of the WH approximation using a single K. For BH we obtain K(ls, is) = 2.16, K(ls, 2s) = 1.38, K(ls, 2p) = 1.69, K'(ls, 2s) = 1.88, and Kave = 1.82; and for BB we find K(ls, is) = 1.35, K(ls, 2s) = 1.89, K(ls, 2p) = 2.05, K(2s, 2s) = 1.26, K(2s, 2p) = 1.88, K,(2p, 2p) = 1.65, Kw(2p, 2p) = 4.89, K'(ls, 2s) = 1.87, and Kave = 1.97, where primes indicate that both orbitals are on the same center. First, we note that the core approximation neglects appreciable off-diagonal terms because the s orbitals on a given center are not orthogonalized; consequently, poor results are obtained. Second, the HF a's8 differ significantly from extended Huckel VSIP's. Third, there is significant scatter in K. The low values of E2" of BH and E3m of BB may be attributed to the small values of K(ls, 2s) and K(2s, 2s), respectively. Fourth, a small but appreciable error also arises from the fact that the one center ls-2pu and 2s-2po- pairs have nonzero /3's, even though corresponding overlaps are zero. Furthermore, the large values found for K. here and in other diatomics9 suggest that wherever a or--r separation is feasible (aromatics, diatomics, etc.), a different K, and K, should be used, and that present computer programs should be modified to allow a different K for each type of overlap situation. We attribute the remarkable agreement of the extended Huckel results with the HF calculation to a fortuitous cancellation of errors introduced by use of the core approximation, incorrect ai's, and a constant value of K. Improvements of each of these types of errors are within reach at this time, and are being explored. Since a good approximation to the correct HF Hamiltonian matrix should yield reasonable eigenvectors and eigenvalues, we may hope to obtain useful chemical information from the wavefunction and energy levels, even in cases where the approximation that A is small is too coarse to yield sufficiently accurate binding energies. Meanwhile, we hope that future publications of molecular HF calculations will include the Hamiltonian matrix elements over atomic basis orbitals. It is to be emphasized that we are not attempting to make the extended Huckel theory compete with SCF procedures, but rather we are using the latter to calibrate the extended Huckel method for use in areas where SCF calculations will not be feasible for some years to come.
* We wish to thank the Office of Naval Research for financial support. 1 For recent examples. see (a) Hoffmann, R., and W. N. Lipscomb, J. Chem. Phys., 36, 2179 (1962); 3489 (1962); ibid., 37, 2872 (1963); (b) Lohr, L. L., and W. N. Lipscomb, J. Chem. Phys., 38, 1607 (1963); (c) Hoffmann, R., J. Chem. Phys., 39, 1397 (1963). 2 M. Wolfsberg and L. Helmholz [J. Chem. Phys., 20, 837 (1952)] approximate the off-diagonal Hamiltonian elements jij as (K/2)(ai + aj)Sij, where ai and aj are diagonal elements and Sij is the overlap matrix element. 3 The diatomic wavefunction program of R. K. Nesbet [Rev. Mod. Phys., 35, 552 (1963)], extensively revised by R. M. Stevens, has recently provided us with the desired matrices. Calculations were done with Clementi's best-atom single-zeta basis set. See Clementi, E., and D. L. Raimondi, J. Chem. Phys., 38, 2686 (1963). 4 Slater, J. C., Structure of Molecules and Solids (New York: McGraw-Hill Book Co., 1963), Vol. 1, p. 106. See also ref. 1(c). 5 (a) Ransil, B., Rev. Mod. Phys., 32, 239 (1960); (b) Clementi, E., J. Chem. Phys., 36, 33 (1962); (c) Woznick, B. J., J. Chem. Phys., 40, 2860 (1964); (d) Pitzer, R. M., and W. N. Lipscomb, J. Chem. Phys., 39, 1995 (1963) (staggered conformation); (e) Moscowitz, J., and M. Harrison,

(3j

VOL. 52, 1964

CHEMISTRY: NIKOLETI(, BORYI(, AND SUNKO

893

M.I.T. Quart. Prog. Rept., 50, 11 (1963); (f) McLean, A. D., J. Chem. Phys., 32, 1959 (1960). 6 Clementi, E., C. Roothaan, and M. Yoshimine, Phys. Rev., 127, 1618 (1962). 7K = 1.75, a(2s)= -0.559 au; a(2p)= -0.312 au; a(H) -0.50 au. 8 For BH a(2s) = -0.909 au; a(2p) = -0.314 au; a(H) = -0.591 au. For BB a(2s) -0.920 au; a(2pr) = -0.189 au; a(2pir) = -0.179 au. 9 Newton, M. D., unpublished work. In all diatomics studied (CO, C2, B2) K7r was quite large. In C2 38 (internuclear distance that of ethane) Keg was 6.35, whereas in staggered ethane [Pitzer, R. M., and W. N. Lipscomb, J. Chem. Phys., 39, 1995 (1963)] the same pair of orbitals (which can now interact with E' and E' symmetry orbitals formed from the hydrogens) had a relatively normal Kay of 2.31. It is tempting to speculate that in larger molecules where full valency is used, the Kay anomaly will disappear.

SECONDARY DEUTERIUM ISOTOPE EFFECTS IN SOLVOLYSES OF SMALL-RING COMPOUNDS* BY MARIJA NIKOLETIC, STANKO BORCIC, t AND DIONIs E. SUNKO
RUDJER BOSKOVIC INSTITUTE, ZAGREB, YUGOSLAVIA

Communicated by John D. Roberts, August 31, 1964

The importance of nonclassical ions as relatively stable intermediates in SN1type reactions is a matter of considerable current controversy.' A very important question in this controversy is the extent to which the observed rate accelerations are due to nonclassical electron delocalization in the transition states as compared to high initial state free energies because of steric strain. The cyclopropylcarbinyl system is one of the numerous examples where high reactivities have been attributed to nonclassical stabilization of the transition state. The cationic reactions of this system have been explained in terms of bicyclobutonium-ion-like intermediates (I).2
CH2-CH 1,,- 1'' +-I CH2---CH2 I

Although a considerable amount of experimental data could plausibly be correlated in terms of intermediates like I, the peculiar behavior of this system is still not fully understood. Thus, phenyl substitution on the cyclopropane ring does not cause any very large increase in solvolytic reactivity.3-5 This marked insensitivity of the rate to phenyl substitution on any other but the a-carbon has also been observed in solvolyses of other compounds where formation of nonclassical ions has been postulated and has been used as an argument against the intermediacy of such ions.6 7 In continuation of our work on secondary deuterium isotope effects,8, 9 compounds II and III have been prepared and their ethanolysis rates measured.

><CH20SO2Me
II

>~CD3

CD3

CD)3,

CH2OSO2Me
III