Volume 4 Editor-in-Chief Paul von Ragué Schleyer University of Georgia, Athens, USA, Professor Emeritus Universitat Enlangen-Numberg, Erlangen, Germany ENCYCLOPEDIA OF COMPUTATIONAL CHEMISTRY ee JOHN WILEY & SONS Chichester - New York - Weinheim + Brisbane - Singapore - Toronto2574 SEMIEMPIRICAL METHODS: INTEGRALS AND SCALING 25. D.G.Oakenfull and DE, Fenwick, Phys. Chem. 1974, 78 1759-1763, 26, V-Gogonea and E. Osawa, 303-317 27.1. P.M, Postma, H. 1. €, Berendsen, and J. R. Haak, Favaday Symp. Chem, Soe, 1982, 17, 95-67, 28. W.1. Jorgensen, J. Chandrasekhar. J.D. Madura, R.W. Impey and ME. Klein, J. Chem. Phys, 1983, 79, 926-935 29, (@)M,L Mandell and H. Reiss, J. Stat Phys, 1975, 13, 107-112; (b) R.M.C. Gongalves, A/M.N. Simies, "and J.J, Moaa-Ramos, J. Soluvion Cher. 1993, 22, 507-517. 30, ©. Sinanogi, “Molecular Associations in Biology", ed B, Pull: rman. Academic Press, New York, 1968. p. 427-445 31.0. Sinanoglo, Theor Chim, Acta, 1974, 33, 279-284 32. M.H Abraham and A Nasehzadeh, J. Chem. Soc. Faraday Trans, 1, 1981, 77, 321-340. 33, (a) C. B, Goganea, “MS Thesis, Toyohashi University of Tech nology, Toyohashi, 1996; (b) C. B. Gogones, V. Gogonca, and E, Osawa, in "Prot, Int Chem, Conf, Pacific Basin Soe, Hon clu, Hl, 1995, pp. 17-22. 34, P.Claverie, “Intermolecular Interactions: from Diatomies 19 Biomolecules’ ed, B. Pullman, Wiley. Chichester, 1°73 35, V. Gogonea, “PHD Thesis’, Toyohashi University of Technol ‘fy, Toyohashi, 1996 36. (a) V. Gogones, B.C, Gogones, and E, Osawa, in “Proc. Kyushu Ine, Symp. on Physical Organic Chemisty’ KISPOC: VE, Fokwoko, Japan, 1995, pp. 328-133: (b) V. Gogonea and E. Osawa, J Comput, Chem, 1995, 16, 817-842. 37.1.0. Hirschfelder, C.F. Cutss, and R. B, Bird, “Molecslar ‘Theory of Gases aud Liquids’, John Wiley, New York, 1964. 38. (a) P.Th. var Duijnen. A, H. Jaffer, and H. Dijkman, J ‘Mol. Sic (Theochem), 1982, 260, 195-205: (b) D. Rinaldi BJ Costa-Cabral, and J-L. Rivail, Chem, Phys, Lert, 1986, 225, 495-499; (c) M. A. Aguilar and F. J. Olivares del Valle, Chem, Phys. 1989, 138, 327-336, 39. (@) MJ. Huron and P. Claverie, J Phys. Chem. 1974, 78, 1862-1867: (o) JoL. Rivail and D. Rinale. Chem. Phys. 1976, 18, 233-242; (©) O. Tapia, F. Sussman, and E. Poulain, J. Theor. Biol, 1978, Th 4 40. W.C. Sill, A. Tempoayk, R. C. Hawley, and , Hendrickson, J. Am. Chem. Soe, 1990, 112, 6127-6128. 41, C5. Cramer and’D. G.Trublar, "Reviews in Computational Chemisty’, eds. KB. Lipkowitz and D. B. Boyd, VCH, New York. 1995, Vol. 6, pp. 1-72. Chem, 1994, Supramol Semiempirical Methods: Integrals and Scaling James J. P. Stewart Stewart Computational Chemistry, Colorado Springs, CO, USA 1 Inwoduetion 2s 2. The Roothaan-Heall Equation 2st 3 MINDOD Semiempircal Approximation 2575 4 1 Approximation 2515 5 128 of the NDDO Formalism 2377 6 ns of NDDO Formalism, 2317 7. Related Anicles 2578 8 Reterences 2518 1 INTRODUCTION ‘Modern semiempirical methods involve the generation of a self-consistent field. In this, the electron distribution is made 4 function of the kinetic and potential energy of the electrons, In part the potential energy depends on the distribution of the clectrons: that is the energy of the electrans is a function of the interelectron imeractions. In LCAG methods, these inter. actions are limited to two-electron interactions, as described by the Roothaan equations.! Because evaluation of the two-electzon interaction energies is very difficult, the ee velopment of SCF methods has evolved inwo wo branches, one oF which, the ab initio methods, attempt the rigorous and nonempirical evaluation of these terms. The other branch, semiempirical methods, avoids even attempting the evaluation of the integrals involved; instead they are replaced with approximations. ‘The purpose of this article is to describe the two-electron approximations used in the NDDO? series of semiempirical methods, MNDO (see MNDO), AMI (see AMI3,* and PM3 (see PM3)3 and to discuss the implications of these approx. mations on the ability of the resulting methods to model systems of chemical interest, 2. THE ROOTHAAN-HALL EQUATION Semiempirical methods, like ab initio methods, define the elements of the Fock matrix, Fy, using the Roothaan-Hall, equation," as shown in equation (1), Fi, Hawt DY [Pe uvide} = Phgluatves] (1) in which Hy« is the one-electron Hamiltonian, PZ, is the density matrix, and (jv)ao) are the two-electron integrals over stomie orbitals, #8 shown in equation (2) twrvor= [ [> ote g vet dn ae @ For the purposes of the discussion to follow, the Rootiaan ‘equation in its spin-free form (equation 3) is more convenient to use, Serourive)] G) Hy $Y [Pieno)— ‘This equation is difficult to solve for several reasons, Most important, the evsiuation of the two-electron integrals is, extremely difficult, {n ab initio methods, the atomic orbitals are approximated asing a Gaussian series expansion. This allows, to be evaluated to any desired level of confidence, but ie computational effort is still great, and, for large systems, this approach is prohibitively slow. Nevertheless, such methods are capable of very high accuracy. Semiempirical methods followed a different path, Instead of trying to solve the imegrals, they are replaced by approx- imations. This resulted in the Uevelopment of many closely related yet distinct methods, the more popular of which were CNDO.#4 INDO methods MINDO/3!® and SINDO," the NDDO methods, and ZINDO. All these methods use a com- ‘mon approximation, that is, all two-electron three- and four center integrals are set to zero, Because there are differencesbetween the various semiempirical methods, the following dis- cussion will be restricted to the MINDO/S and NDDO approx- ‘mations. These illustrate the theory involved, that i, with only ‘minor modifications, she ideas developed by the discussion of these methods can be applied to the other methods. 3. MINDOG SEMIEMPIRICAL APPROXIMATION ‘MINDO/S was the frst of the modem semiempisical meth- ods. The main improvement over earlier methods was the use ‘of molecular rather than atomic data in determining parame- ters, In this method, the two-electron integrals were partitioned ‘nto two distinct types: one-center integrals, those dependent fon only the type of element involved; and two-venter inte grals, which depended on the types of atoms involved and on the interatomic separation. 3.1 One-center Integrals In MINDOV3, the values of the one-center twovelectron integrals were obtained from atomic spectra. For atoms with an s-p basis set, there ae six one-center two-electron integrals, as shown in Table 1. One of these integrals, however, is related to two others, as shown in equation (4); consequently there are only five independent one-center integral, Hye = (Gap ~ Gop) a In performing semiempirical calculations, it is essential that ‘his relationship be maintained: if it ig not, then rotational invariance, an essential for any valid method. would not be maintained, Because these quantities are atomic constants, they should not be regarded as being parameters. 32 Two-cemer integrals ‘The approximation to the two-center (woreleetson inte tused in MINDO/3 is that proposed by Dewar and Sabel and is shown in equation (5), 4 e 6) in which expfession ¢ has the value 14.397, as indicated in equation (6. o Table 1 One-center Tw0- leczon Integrals Integral ‘Symbol {ssis) (ssipp) (isp) ‘wri, ‘pipe (op'pe> IPIRICAL METHODS: INTEGRALS AND SCALING 2575 The properties of this function are of interest. For very somal Ry, the function cottapses to yay. That is, the function correctly predicts the one-center two-electron enesgy. At very lsge distances, the function collapses to 1/Ras, which is the expected behavior for two point charges. At imermediate distances, the function is slighily less than that expected by the point-charge approximation, 3.3. Limitations of the yay Approximation ‘This function depends on the properties of the two atoms involved, but does not differentiate between the various atomic otbitals involved. As a result, the electostate interactions of 4p orbital are the same as those of an s orbital. While this keeps the method simple, it does reduce the flexibility of the method, “More importantly, the funetion contains no terms whieh can model the behavior of a lone pair because all integrals of the type (vagelgnpe) in which 2 is 90 the same as are zero, lone pairs, which involve s-p hybrid orbitals, cannot be represented ‘A consequence of this is that phenomena that depend on lone pair-charge or lone pair-Ione pair interactions cannot be rondeled. Ths situation cansot be rectified by an appropriate choiee of parameter values ~ the model simply dees not have the flexibility to allow such phenomena so be modeled, 4 NDDO SEMIEMPIRICAL APPROXIMATION In 1977 the NDDO appreximation® was proposed. In this, many of the defects of MINDO/S were corrected. In partic lular, the two-electron two-center integral approximation was cchariged to allow all integrals involving any two orbitals on one tom with any two orbitals on a second atom to be modeled This increased the aumber of integrals dramatically. For each atom with an s-p basis set there are 1G distinct pairs of atomic orbitals (Table 2), Therefore, for wo dissimilar non-hnydeogen atoms, thete are a total of 100 different two-electron two-center integrals, where in MINDOY3 there had been only one. ‘The NDDO approximation for the new integrals can best be understood in terms of the point-charge expansion of the atomie orbital pairs. Examples of the four possible multipolar charge distributions are shown in Figure 1. The s-s charge distribution is spherically symmetric, and therefore can be represented by a single unit charge at the atom. A lone pair is represented by the s-p distribution, This has a net charge of zero, and behaves like # dipole centered on the atom, Bach p-p pair behaves like the ©s pair, but is modified by the angular dependence of the orbital. Like the s-s distribution, the net charge is unity. Finally, the p-p' distribution corresponds to a sivuple quadrupole. yy 001 o toonin = oT Trametes) ‘Fable 2 Atomic Orbital Pairs sp Pa Spo PP ay se pye PD2576 _SEMIEMPIRICAL METHODS: INTEGRALS AND SCALING “ “ ve we | 40 a ~ eo By ve we rer sess —e @ eo 3: Figure £ Point charge distributions for pairs of etomie oxbitals| ‘The general form of the approximation for the two-electron two-center imegral is as shown ia equation (7), in which Rys(i) is the distance between point-charge Q, on atom a and point charge Q; on atom b, and the d quantities are functions Of the atomic orbitals on the respective atoms. ‘To see how these quantities relate, consider the integral (yp,lsp,), involving two atoms, as shown schematically in Figure 2. The six distances Ryp(ij) then represent the distances between cach of the three’ point charges representing the PP, distribution and the two point charges representing the sp, distibution, In this approximation, the quantity Dy is, chosen s0 that the multipole moment due to the point charges reproduces that expected fiom the produet of an s and p atomic “orbital, and can be calculated from the orbital exponents using equation (8). an tabs a ED eg AB Gt oY A similar expression exists for Dz, based on the p orbital ‘exponents (equation 9), Qn + 1)@n +2) ye [SED » Unfortunately, the definition of the quantities dy, is not so straightforward. In MINDO/3, the equivalent quantity was chosen so that the integral would collapse to Gy, aS Ray Went to zero. While this is the same in the NDDO approximation Figure 2 Two-tecrom inca (9,962 when the pair of orbitals are of type s (equation 10), the values wher ane or both orbitals are of type p is more complica, Aig = oy ‘The nature of the problem when one or more p orbitals are involved can be seen when an attempt is made to evaluate the dy term, The inceractions involved are shown in Figure 3 ‘There’ are four of these: between the -+e/2 and —e/2 point charges on atom a and the +e/2 and ef? point charges on atom b, At Ryy ~0, the equation for the integral is given in ‘equation (11), in which the distance Ryayu = yy =0. and Raya = Raya = 2D ay ‘This expression can be re-arranged to give a quartic equa. tion (equation 12), Dictdt, — 4H Died), + 4H2,D}d2, ~ Hycdsy + Hy, a2) No analytic solutions exist for such equations, but they can be solved by iterative means to any accuracy desired, OF the four solutions, only the real positive solution is meaningful By inspection of the numerical values of the constants in equation (12), @ stating estimate for the valve of ty cae be ae en ‘, ¢ oe) «a ‘This estimate can be improved by use of the first differential (eguation 14). After four nr Sve iterations, the error in value ‘of day becomes vanishingly small Died, ~ 4H pDied HHi03 ORT — Hegey + : 2 — 1 Died + BHEDHdiy ~ Hage «sy ven +0 a of Figure 3. The (sp[sp)interetionSEMIEMPIRICAL METHODS: INTEGRALS AND SCALING 2577 “Table 3 Parameters for Carbon Parameter Valve Cale Quantiy Value Gu 231223 Di 0.807466 Gm a7 1080 Dy 0.685158 Gr 108 1038 dy 0.440467 2 98s 10.00, ty 0614937 Hy 2433 on 0.668590, He 062062 A similar, although slightly more complicated, expression can be derived for dp, AUR =O, the expression far dp is given by ‘equation (15), in which the quantity Ds is the distance from fan axis (0 a point-charge, as seen in the p-p’ interaction in Figure 1 © a, 1)? : ‘The expansion of this term to form a polynomial is tedious. Instead, the value of dp can be obtained directly using 4 starting value and iterating using equation (15) and its derivative, as shown in equation (16). slo-slena) in which the starting value of dyy is given by equation (17), : inn 0s(te)? wo ‘Convergence is achieved in four to five iterations ‘The advantage of using an approsimation of this type is that no new parameters are inttodvced into the method. Inthe NDDO formalism, all wwo-center eworelectton integrals are simple functions of the atomic osbitas, the one-center two electron integrals, and the relative positions of the two atoms Toth the one- and two-center integrals used in semiempiri cal methods are smaller than the equivalent values predicted from a fst principles or analytical calculation (Table 3). This has been attributed! to the implicit inclusion of correlation effects in the semiempirical values. 5 ADVANTAGES OF THE NDDO FORMALISM A consequence of the inereased sophistication of the NDDO ‘method over that in MINDO/3 is that more subtle phenomena can be modeled. Probably the moet important specific improve- ‘ment was that the electrostatic properties of lone pairs were nowy correcily modeled. The overall effect was that, for almost all types of systems, the new method gave much improved results 6 LIMITATIONS OF NDDO FORMALISM ‘The NDDO formalism has very few limitations. Most che- rmical properties ean be related fo atomic or interatomic inter- actions. ‘These interactions are accomed for in the NDDO formalism, so such phenomena can be modeled by this method. Very few, if any, phenomena can be related 10 titomic or four-aton jnteaetions. These are the types of interaction that are represented by ab initio caleulations, and are ignored in semiempirical methods. Therefore the fac that these intrac~ tions are ignored does not provide prima facie evidence of a limitation in these methods. ‘There is one fault in the NDDO approximation to the Wwo- electron integeals, « fault shared by the simpler approxima- tion used in MINDO/3. This fault does not manifest itself in caleulations on molecules or ions, or in calculations on poly ters or layer systems, or even calculations on solid elements. Strangely enough it only shows up in calculations of thee- dimensional solid-state compounds. To understand the nature of the problem, consider a solid compound of formula AB. W atom A caries a charge Q, then atom B caries @ charge of =O, and the potential generated at an atom A due to the two atoms in a distant unit cell will be of the form shoven in equation (18. & [ie (aa "aa -7_ as eat (241) ea a * daw Here the multipole expansion has been ignored: at large distances each atom behaves like a point charge. This potential thas the form shown in equation (19), Vas Yaa Oa (he tert a+?) avy ‘of, on expanding in a power series, Qalla =) vga & OR, 20 ‘The significance of this expression is that the potential generated at a point by {wo equidistant atoms eamcying equal and opposite charges is not zery. The difference is, admittedly, small, and for discrete species, polymers and layer systems, the error is negligible, For solids, however, ths is not the ease. For solids, the number of unit cells at any given distance from an atom rises as 4, therefore the potential duc 10 shell of unit cells at distance R is as shown in equation (21), en2578 SEMIEMPIRICAL METHODS: TRANSITION METALS. This is still a negligible quantity. But the potenti) at an atom is due to unit cells at all possible radii, and that sum is, infinite (equation 22), Vana trot to} Va =00 a ‘The implication of this is that the calculated semiempisical potential experienced by ar acom in a solid compound would be infinite. This is likely to hinder the development of a useful semiempirical solid state method. 7 RELATED ARTICLES AML; Integrals of Electron Repulsion; Integrals: Over- lap; MNDO; MNDOMd; PM3, 8 REFERENCES 1. CC. Roathaan, J. Rev. Mod. Phys. 1951, 23, 69-89 2. IA PopleandD. L. Beveridge, “Approximate Molec tal Theory’, McGraw-Hill, New York, 1970. 3. M.J.S. Dewar and W. Thiel, J. Am. Chem. Soe, 1977, 99, 4899-4907, 4. M4. S. Dewar, 6, G. Zocbseh, E, F. Healy, and J.J. P. Stew ‘at, J, Am Chem: Soc., 1985, 107, 3902-3909. 5. J.1.P. Stewart, J, Comput Chem , 1989, 10, 209-220. 6 G.G. Hall, Proc. R, Soc, London, 1981, A208, 541-542 7. 5.8 Pople and G.A.Segal, J Chem Phys, 1965, 43, 8136-8149. 3. J.A.Pople and G. A. Segal, J. Chem 3289-3296, 9. 1A. Pople, D. L. Beveridge, and P. A. Dobos, J. Chem, Phys, 1967, 47, 2026-2083. 10. RC. Bingham, M. J. 8. Dewar, and D. H. Lo, J. Am. Chem Soe,, 1975, 97, 1285-1293 11. DLN. Nanda and K. Jug, Theor. Chim. Acta, 1980, $7, 95-106. 12. L, Oleari, L: DiSipio, and G. DeMichels, Mol. Phys, 1986, 10, 9-109, nb Phys, 1966, 4, 13. M.J.S. Dewar and N. L.Sabelli, J. Chem. Phys, 1952, 66, 2310-2316, 14. MJ.S. Dewar and 9, Thiel Titer Ohm, Acta, 1977, 46, 89-104, Semiempirical Methods: Transition Metals Andrew J, Holder Universipy of atssouri Kansas City, Kansas City, MO, USA 1 Discussion 2578 2 Related Anicles 2519 3. References 2579 Abiyreviations MNDO = modified neglect of differential overlap, MNDOMd = modified neglect oF differential overlap with d- functions. 1 DISCUSSION ‘Transition metal elements are a real test of the efficacy of quantum chemical theories. Primarily, this is because many’ (Gf noe most) of these systems have unpaired electrons, i.e. are open shell, Open shell systems therefore require additional approximations beyond standard Hartree-Fock theory co make the calculations tractable computationally. These include the UHF (unrestricted Hartree-Fock) and configuration interaction (CD) approaches. Both approaches are computationally costly and have complicating considerations that make their general application more difficult and less straightforward, Also, itis, often difficult to determine which spin state is the correct ‘of natural one, and properties can vary substAntially with differences in spin state ‘A few semiempirical approaches are currently available that may allow treatment of transition metals with efficiency. ‘Two of these fall in the INDO class of HF models, (INDO methods are usually more computationally efficient, though not as accurate, as those based on NDDO theories.) INDOVS? developed by Zemner af is included in the ZINDO? package and has been available for a number of years. It is. very successful for the prediction of electronic spectra, for which it ‘was carefully parameterized. However, INDOVS is not usually applied to compute more general properties such as optimized geometries or molecular energies, a8 it is not deemed 10 be reliable for these values? SINDOI* from Jug et al. has recently been expanded to some transition metal elements and performs as expected within the limited INDO theory used, Again, results are somewhat erratic and the method does not ‘appear to treat open-shell systems with 2 great deal of success. ‘There are currently four NDDO methods for transition met- als, only thtee of which have found some acceptance and use. SAMS MNDOVS and PM3(imn)? ae all implemented in commercial software packages (AMPAC® Unichem,? and Spartan!” respectively) that are commonly gvailable. Filatov et al. have also developed a method they call NDDO/MC2! ‘which was published in 1992, While the three more accepted methods ate intended to be general models predicting a wide range of chemical properties as the previous Dewar-style meth: ods have done, NDDO/MC was aimed at reproducing geome- ies and binding energies, a which it performs relatively well. It was thus not directly parameterized to compute such items as heals of formation and electronic properties. Values Of this type are needed for the study of reactions, an area in Which semiempirical methods have traditionally been used the most extensively, NDDOUMC witt find usiity in certain cit- cumstances, but will probably not become a general purpose ‘method. ‘The MNDO/d method from Walter Thiel et al, was also. Introduced in 1992, This method still utilizes a modified point: charge model (multipole expansion ax in AM? ancl PM3) for computing the two-center two-electron integrals (TERIS). MNDOvd is essentially layered onto the older (1977) MNDO. model for the elements where Z < 11. Heavier elements Were then ro-parameterized utilizing an expanded version of