
Citation 
 Permanent Link:
 http://ufdc.ufl.edu/AA00030049/00001
Material Information
 Title:
 A quantum chemical approach to the determination of the spin distribution within large molecular radicals
 Creator:
 Cory, Marshall George Jr., 1955
 Publication Date:
 1994
 Language:
 English
 Physical Description:
 ix, 133 leaves : ill. ; 29 cm.
Subjects
 Subjects / Keywords:
 Approximation ( jstor )
Eigenfunctions ( jstor ) Eigenvalues ( jstor ) Electronics ( jstor ) Electrons ( jstor ) Hydrogen ( jstor ) Magnetism ( jstor ) Molecular orbitals ( jstor ) Orbitals ( jstor ) Wave functions ( jstor ) Chemistry thesis Ph. D Dissertations, Academic  Chemistry  UF Quantum chemistry ( lcsh ) Quantum theory ( lcsh )
 Genre:
 bibliography ( marcgt )
nonfiction ( marcgt )
Notes
 Thesis:
 Thesis (Ph. D.)University of Florida, 1994.
 Bibliography:
 Includes bibliographical references (leaves 124132).
 General Note:
 Typescript.
 General Note:
 Vita.
 Statement of Responsibility:
 by Marshall George Cory.
Record Information
 Source Institution:
 University of Florida
 Rights Management:
 Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for nonprofit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
 Resource Identifier:
 030547072 ( ALEPH )
31743901 ( OCLC )

Downloads 
This item has the following downloads:

Full Text 
A QUANTUM CHEMICAL APPROACH TO THE DETERMINATION OF THE SPIN DISTRIBUTION WITHIN LARGE MOLECULAR RADICALS
By
MARSHALL GEORGE CORY, JR.
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1994
To
my beloved mother Catherine, my father Marshall Sr., his
wife Jeannie and my dear wife, Genny
ACKNOWLEDGMENTS
It is impossible to acknowledge all who have contributed to my University of Florida experience. With intentions to slight none, I choose to mention a few who "stick out" for one reason or another.
* Dr. Michael Zerner, for his constant enthusiasm.
* Dr. Charles Taylor; Charlie, the "a la personne qui a garde la porte du club au
moment opportun" would not have been necessary if you and one other had not ran
and hid like "wimps".
* Dr. Ricardo Longo, my best friend at QTP and the best read graduate student I know
of, or likely ever will know of.
* Genny Cory, for putting up with this nutty endeavor over the years.
Finally to all the friends I can remember at this moment, Rajiv and Priya, Ricky and Ivani, Quim, Steve and Trish, Christin, YaWen, Billy Bob, Chris, Mark, Dave, Zhengy, and of course Monique. To the small multitude of others, whose friendships I have enjoyed and whose names escape me at this moment, forgive me for not remembering in time.
11i
TABLE OF CONTENTS
ACKNOWLEDGMENTS ................................. iii
LIST OF TABLES ........ . ............... . ................ . vi
LIST OF FIGURES ..... ...... ......................... vii
ABSTRACT ........................................ viii
CHAPTERS
1 INTRODUCTION ................................ 1
The ESR Experiment .................................. 1
The Isotropic Hyperfine Interaction ................. .... 5
Overview of SelfConsistant Field Theory .................. 10
The BornOppenheimer Approximation .................. 12
The Independent Particle Approximation ....... .......... 13
The INDO Approximation .......... ...... ... ......... 25
2 THE UHF APPROACH ............................. 32
Introduction ... ............. ... ..................32
The UHF Equations ........ ....................... 33
Spin Contamination .................................. 37
Other Interesting Orbitals ................................. 39
Spin Contamination and the UHF PE Surface .............. 44
Spin Projected UHF ......................... .... 46
The AUHF and EUHF Methods ................... ...... 55
Calculated Results for the UHF and PUHF Methods. ............ 57
3 THE ROHFCI APPROACH ......................... 61
Introduction ..... ............................... 61
Restricted OpenShell HartreeFock ..................... 61
Configuration Interaction .............................. 63
Correlation Energy ................ ................ 65
Single Excitation Configuration Interaction ................. 67
iv
SpinAdapted Configuration Functions . . . . . . . . . . . . . . . . . . . . 74
Rumer Bonded Functions ............................. 74
Evaluation of pZ for a Rumer CI Wavefunction . . . . . . . ...... . . 78
Fast Doublet CIS ................................ 81
FDCIS SpinAdapted Configurations ..... .............. 81
Formulae for the Matrix Elements of DCIS . . . . . . . . . . . .... 83
Calculated Results for the ROHFCIS Method . .............. 85
4 CONCLUSIONS .................................. 88
APPENDICES
A UNITARY ORBITAL ROTATIONS ............................. 95
B THE MIRROR THEOREM ............................. 97
C THE PAIRING THEOREM .................... ... 100
D THE EVALUATION OF I Ons(0)12 ..................... 103
The Optimum Slater Valence Exponent .................. 108
The Charge Dependence of ................................ 114
The Results of Optimizing the Valence Exponent ............. 117
REFERENCES ................. ..................... ..124
BIOGRAPHICAL SKETCH .............................. 133
V
LIST OF TABLES
Table 21: The affects of annihilation on (S ) and calculated ai .............. 55
Table 22: Calculated results for Bchla. .. ............... . . . . . . 58
Table 23: Calculated results for Bchlb and Bphea. . .................. . 59
Table 31: How the correlation of the INDOROHF reference affects the calculated value of aiso .................................... . . 72
Table 32: Calculated results for Bchla and Bphea . . . . . . . . . . . . . . . . . . . . 86
Table 33: The calculated results for Bchlb . ................. . . . . . 87
Table Dl: The (os of the standard Nitrogen optimization data set, where ~, = 1.95. 108 Table D2: The (4 of the standard Hydrogen optimization data set, where ( = 1.20. 109 Table D3: Minimization results for rhydrogen . . . . . . . . . . . . . . . . . . . . 113
Table D4: Minimization results for ohydrogen . . . . . . . ...... . . . . . . . . . 113
Table D5: Minimization results for rnitrogen . . . . . . . . . ........ . . . . 113
Table D6: The optimized hydrogen exponents ......... . . . . . . . . . . . . . . 114
Table D7: The optimized nitrogen exponents . . . . . . . . . . . . .. . . . . . . 114
Table D8: Computed UHF, PUHF and ROHFCIS results for the optimum Slater valence exponent ....... ..... .......... ... ........ 115
Table D9: The effects of including the two and threecenter contributions to a,,,. 116 Table D10: The Optimization Data Set ...... .................... 117
vi
LIST OF FIGURES
Figure 1: The electron magnetic resonance condition ................... 3
Figure 2: Rumer bond diagrams for the case of three electrons in three orbitals and
S= . ...................................... 77
Figure 3: A closed loop and an odd chain. ................... ... 78
Figure 4: Two types of even chains which arise from (1 2) interacting with (1 (2 in
the first case and (1 2) (3 4) interacting with (1 (2 3) (4 in the second. . 78 Figure 5: A multicycle closed loop, arising from the interaction between (1 2) (3 4)
and (1 (2 3) 4). ............................. .... 79
Figure 6: The four possible orbital configurations for the FDCIS formalism..... 82 Figure 7: The Nitrogen is probability density. ................... . 119
Figure 8: The Hydrogenlike 2s probability density. .................. 119
Figure 9: The Slater type 2s' probability density . .................. 120
Figure 10: A plot of the spectroscopic CIS P8,, vs the experimental Ao. ..... 120 Figure 11: The ir region of Figure (10) expanded. .............. . . . . . 121
Figure 12: A plot of UHF ac.x, vs a,,, using (2,= 0.961. .... .. ......... 121
Figure 13: A plot of PUHF a,,e vs a,,,, using (2,= 1.389. .... ........... 122
Figure 14: A plot of aC..C vs a,,p, using (2,=1.248. ............. . . . . . 122
Figure 15: The number system for the Bacteriochlorophylls ........ . . . . .. 123
vii
Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
A QUANTUM CHEMICAL APPROACH TO THE DETERMINATION OF THE SPIN DISTRIBUTION WITHIN LARGE MOLECULAR RADICALS By
MARSHALL GEORGE CORY, JR.
April, 1994
Chairman: Michael C. Zerner
Major Department: Chemistry
Within the last decade the anticipation of some experimentalists for the ability to ascertain the structure of biologically important chemical systems has been realized. The elucidation of the structure of the photosynthetic reaction center of R. Viridis, including some 13,000 atomic centers, is a prime example, and this structure has led to a deeper understanding of the mechanisms of photosynthesis. Along with this increase in structural knowledge a concurrent increase in the knowledge, of the observable properties of such structures has also occurred. Of particular interest to this work are the magnetic properties of these systems, as revealed by Electron Spin Resonance (ESR) spectroscopy. The magnetic spectra, ESR, of such systems are quite complex and require some effort to understand, and a need for theoretical aids has been expressed. In many cases the ESR viii
signal is the only structural signature available.
This work has explored several quantum chemical methods for determining the isotropic hyperfine splitting (hfs) in such systems as those described above. Several semiempirical SelfConsistent Field (SCF) and postSCF approaches are explored. Two distinct parameterizations of the Intermediate Neglect of Differential Overlap (INDO) approximation are examined for their ability to produce good reference functions. Both Unrestricted HartreeFock (UHF) and Restricted OpenShell HartreeFock (ROHF) SCF wavefunctions are investigated, as well as the Configuration Interaction Single replacement (CIS) wavefunction. The results of applying the most promising method to several of the bacteria chlorophyll ions are reported and compared to those observed experimentally.
ix
CHAPTER 1
INTRODUCTION
The ESR Experiment
In general, magnetic resonance spectroscopy observes transitions between energy levels which depend on the strength of a magnetic field [13]. Experimental techniques which measure the interaction between an external magnetic field and the intrinsic electron spin magnetic moment [4, 5] are usually labeled electron spin resonance (ESR) spectroscopy experiments. Techniques which deal with interactions between orbital magnetic moments and an external magnetic field are labeled electron paramagnetic resonance (EPR) spectroscopy experiments. These terms are sometime used interchangeably; in this work we shall deal with the ESR problem as defined above.
While any atomic or molecular system containing one or more unpaired electrons will in general possess a magnetic moment due to the intrinsic spin of electrons, we shall be concerned in this work with molecular systems containing a single unpaired electron. (And as we shall see later, really only with those magnetic molecular systems that are "large" in the sense of abinitio quantum chemical theory.) This work is an attempt to develop a systematic and reliable method for the description of the spin distribution within large molecular systems. For radicals whose total electronic spin angular momentum S
1
2
equals 1, or in atomic units' , interactions such as electron spinspin coupling do not affect the experimentally measured spin distribution, and thus do not need to be addressed here and will contribute no fog to the overview of our final results. The vast majority of molecular systems have S values of zero; by far the next most prevalent value for S is 1. Seen in this light then, our stressing spin onehalf systems is not as constraining as it might seem at first, and if our attempt at developing reliable computational methods is successful then such multielectron interactions can be addressed at our leisure.
Let us start with the simplest molecular ESR experiment. Consider a molecular system with a nonzero net electronic spin. Associated with this spin will be a magnetic moment. If a static external magnetic field is applied, the spin magnetic moment of the electron will interact with it. If the total electronic spin angular momentum S of the molecular system is 1, and the molecular system is spatially nondegenerate, then there are two magnetic energy levels available to the electron. These levels are degenerate in the absence of a second magnetic field. The degeneracy is lifted by the application of the static external magnetic field and the splitting is proportional to the external field strength (Ho). The basic ESR experiment then consists of observing transitions between these two levels, or states.
1. In atomic units h = e = me = 1, where h is Planck's constant divided by 2ir, e is the magnitude of the electron charge and me is the electron rest mass. Unless otherwise stated atomic units will be used through out the rest of this work.
3
Transitions between the energy levels can be induced by the application of a second, appropriately oriented, external oscillating magnetic field. In our simple experiment the oscillating field is applied at a right angle to the static external field Ho. Transitions will occur when the resonance condition AE = hvo = gef#eHo is satisfied, Figure (1) below.
E, I hve=:gBH.
Ho
Figure 1: The electron magnetic resonance condition.
Here h is Planck's constant (6.63 x 1027 erg sec), #, is the Bohr magneton (9.27 x 1021 erg/G), g, is a dimensionless constant of proportionality equal to 2.002322 for a free electron and vo is the frequency of the perpendicularly applied oscillating magnetic field. The oscillating magnetic field has the same probability of inducing an upward transition as a downward transition. In a bulk sample at equilibrium in the static field the lower level will be more highly populated than the upper one, thus there will be more upward than downward transitions, and a net energy transfer to the bulk sample. It is
4
this absorption of energy that is detected, or observed, experimentally.
If this was all there is to an ESR experiment then little useful information would be obtained by its being carried out. We would be able to obtain the constant of proportionality g, which would equal the ratio . If for the moment we disregard the possible dynamic information intrinsic in the absorption line width and shape, then this would be the only piece of information we could extract. Fortunately, this simple picture is not complete. The simple system we used as our example above is not to be the case for the vast majority of magnetic molecular systems. Almost certainly there will be other intrinsic magnetic fields within the molecular system. These can be due to other unpaired electrons or orbitals which possess a net angular momentum and probably most importantly, to magnetic nuclei within the system. The interaction between the unpaired electron and the various magnetic nuclei2 gives a wealth of information on the distribution of the unpaired spin within the system; this in turn yields detailed information on such varied subjects as the molecular geometry, the electronic state of the system, the net spin of the magnetic nuclei can be determined, and from this the isotope of the magnetic center being observed. The experimentalist can also detect from the electronnuclear interactions the presence within the system of nuclear spinspin coupling as well as a wealth of other information we need not detail here.
2. A rather complete list of the magnetic nuclei and their various spin values can be found in Appendix (1) of Reference 6.
5
From this brief overview it is obvious that magnetic resonance spectroscopy, for this work ESR, is a powerful tool for sciences' use in the investigation of the invisible world of atoms and molecules. It is just this power to yield, sometimes overwhelmingly so, detailed and complex information on the system being observed that spawned the need for, and our desire to develop, theoretical methods to aid in explaining, and hopefully predicting, the observed spectrum.
The Isotropic Hyperfine Interaction
The approximate molecular Hamiltonian3 is a thirtyfive term construct, some twenty of these terms will influence the experientially observed ESR spectrum. This dissertation shall primarily be concerned with only one of these, the isotropic hyperfine interaction, or the Fermi contact [8], ais,. The origins of the Fermi contact term is explained in terms of a magnetic dipoledipole interaction between the magnetic moments of an electron and a nucleus. The vector potential [9] due to the nuclear magnetic dipole moment is rX (1.1) where we use the standard mks units. The magnetic field due to '4 is
( ) = x (r) = L3  5 (1.2) The interaction energy between IN and ie is
E_ V (r __ _ 3(r  (Aef N)
Edip = [fN  3(e(=  iN  ps (1.3)
3. See Appendix (F) of Reference 7.
6
where f represents a vector joining the electron and nuclear centers, ie and fi, are, respectively, the electron and nuclear magnetic moments. The Hamiltonian for this interaction is written as
Hdip = 9ee9NN IN 3(S r (IN (1.4) r3 r5
where the collection of constants preceding the operators are, respectively, the free electron "gfactor" or Lande gfactor ge, the Bohr magneton #e, the nuclear gfactor gN and the nuclear magneton ,3. The operators S and I are the total angular momentum operators for the electronic system and the nucleus, respectively.
When the electron is found in an orbital with an angular momentum of zero, i.e. I = 0 or an s state, the wavefunction is not zero at the nucleus. (In the Dirac theory [4] the atomic radial function diverges, but the divergence is so small that equation (D.7) is approximately correct for the nuclear dimension.) In any case 7, or 7, of equation (1.3) diverges at the origin.
Suppose we take a small sphere with the nucleus as the center. Then the contribution to the isotropic interaction from the s electron outside the this sphere is zero, since the integral of 1 for this volume stays finite. If the sphere is small enough, the electron density for the s electron inside the sphere can be regarded as a constant, and be replaced by the wavefunction at the origin, i.e. as IJ(0)12. Since the s electron has a magnetic
7
moment ie4, this small sphere is homogeneously magnetized with a magnetization of M = lI(0)l2 (1.5) The magnetic field, Hs, at the center of such a homogeneously magnetized sphere is
g
H, 3 (1.6) or
= 2 81
S= 4r I + M] 4 r(0) (1.7) at the nucleus. The interaction energy between this field and the nuclear magnetic moment is
87r
Eiso =  (Fie ,) (r,)l (1.8) (This is the FermiSegrb term [10] and was first given by Fermi [8].) In operator form equation (1.8) is written as Hso  aisolï¿½ S (1.9) where aiso equals
8r
aiso = "geSN9pz(rN) (1.10) Here pZ(rN) is the unpaired spin density at the center of nucleus (N). As we discussed earlier, an unpaired electron in the presence of a magnetic field has two states available
8
to it, either aligned with or against the field. These states usually are labeled spin up and spin down or as alpha (a) spin and beta (0) spin. The unpaired spin density at a point in space, pZ(r  ro), is then defined as the difference between the a and # spin densities [11] at that point, or as it applies to equation (1.10) above
pZ(rN)  p*(rN)  p (rN) (1.11) The product "!3eg/lN is a constant. The constant of proportionality ge is not altogether constant. Nor can it in general be considered a scalar quantity. It is correctly expressed as a second rank Cartesian tensor [12]. Five terms contribute to the tensorial components [7]. Generally the largest contribution to the anisotropy of g, is through the oneelectron spinorbit interaction [13]
Hso = L.S (1.12) where L is the orbital angular momentum operator and (c) is the spinorbit coupling constant. The oneelectron spinorbit interaction is proportional to the atomic nuclear charge raised to the fourth power, (Z4), [13] and becomes important for the heavier atomic systems. The molecular systems to which the methods developed within this dissertation are to be applied, are themselves constructed from the first, second and occasionally the third period elements. The oneelectron spinorbit interaction in these atomic systems is almost always quite small. The change in the average value of g,
9
for these systems is typically at the third decimal place, an error smaller than the error intrinsic to the computational methods we will use to determine the spin density. Thus, for the methods developed within this dissertation the electronic gfactor will be treated as a scalar constant. This leaves
pZ(rN) (1.13) as the quantity to be determined computationally.
Before describing the general details of how we go about computing equation (1.13) we need to present an overview of the quantum chemical approach we have chosen to use. Our stated objective is to derive a method that can, to predictable limits, determine the unpaired spin distribution throughout biologically important, and usually quite large, molecular systems, and to test it through the calculation of an observable property, as,,. This constraint as well as the desire to start out on known theoretical ground leads one to, as least in my opinion, semiempirical molecular orbital (MO) methods. Semiempirical MO techniques such as the various oneelectron Hiickel methods [14, 15] or the twoelectron ZDO [16] approaches have been applied to such systems in the past [17, 18] with mixed results. Their simple form and low computational cost make HUckel MO methods logical first choices. Collectively they do an adequate job of describing the gross overall spin distribution of systems such as the porphyrins or the chlorophylls, systems that shall concern us later. On the other hand the calculation of quantitatively acceptable results
10
by such methods has not been demonstrated. The assignment of the isotropic hyperfine splittings (hfs) through the use of the McConnell relation [1] and Hiickel spin densities have proven to be qualitatively useful. We expect to do better than this and obtain results that can be quantitatively useful. The aforementioned semiempirical Hickel methods do not treat the electronic twobody interactions of a molecular system explicitly. The ZDO semiempirical methods we choose as our theoretical starting point do. It would seem from the start something of importance is missing from the oneelectron Hiickel methods. Also, primarily for the reason of the ignored electronic twobody interactions the Hiickel methods do not lend themselves to post SCF procedures, techniques we will, in later chapters, find to be of importance in obtaining reasonable values for the spin density, (1.13).
Overview of SelfConsistant Field Theory
Today the abinitio4 molecular orbital theory of choice is the HartreeFock SelfConsistent Field method [1925] (HFSCF). The ZDO theories we have chosen to use as our starting point are themselves of the HartreeFock selfconsistent field type. As such, it is necessary at this point to outline HartreeFock theory before introducing the ZDO methods. It is not our intent to develop the HartreeFock equations in detail as this lies outside the scope of this work. We do however need to give sufficient detail here so 4. Abinitio theories tend to make only those approximations which are necessary to keep the problem soluble.
11
the ground work will have been laid for the methods discussed in later chapters. These methods will address the openshell problem directly. Here we will give an overview of closedshell HartreeFock theory, and leave the explanation of the various openshell approaches to each specific chapter.
What we would like to be able to solve is the nonrelativistic, timeindependent, manyelectron Schri6dinger equation,
H = EqV (1.14) for the wave function IT, where H is the Hamiltonian for the system of electrons and nuclei and E is the energy of the system. Explicitly H is written as
N M NM NN MM
 iF1 ZAZB
H= Vi  V  ZA  +1+ B (1.15)
S A r iA rij RAB
i=1 A=1 i=1 A=I i=1 i
Above RA and ZA are the position vector and charge of nucleus A, and r., the position vector of electron i. The summations run over all N electrons and M nuclei. Term by term equation (1.15) is explained as follows, the kinetic energy operator for the N electrons, the kinetic energy operator for the M nuclei, the Coulombic nuclearelectronic attractions, the electronelectron Coulombic repulsions and lastly the Coulombic nuclearnuclear repulsions.
With two exceptions the wave function WF of equation (1.14) cannot be solved for in closed form. These exceptions are the special cases of 1) the free particle and 2) a
12
two interacting body system, or a system that can be reduced to two interacting bodies. The problem lies in that the motion of the electrons and nuclei cannot be uncoupled for a three or more interacting body system. If we make the following two approximations we can uncouple the motions of the various particles and at least within the validity of the approximations solve for the wave function T. These approximations are known as the BornOppenheimer approximation and the Independent Particle approximation. The BornOppenheimer Approximation
If we invoke the physical picture of the electrons moving among the nuclei so much more rapidly than the nuclei themselves move that we can consider the nuclei to be fixed with respect to the electrons, then we have also invoked the BornOppenheimer approximation [26]. This uncouples the motions of the nuclei from each other as well as from the motion of the electrons. The effect of this uncoupling on the form of equation (1.14) is as follows; the wave function T(r,, RA) now parametrically, rather than explicitly, depends on the nuclear coordinates RA and we can write it as
IF(r, RA) = 4DRA (ri)(RA) (1.16)
where fl(RA) is the wave function describing the now separate motions of the nuclei and
'ORA(r,) = 4el (1.17)
13
is deemed the electronic wave function describing the electrons in the field of the "fixed" nuclei. Treating the nuclear positions as fixed parameters leads to
M
 V = 0 (1.18) A=1
and
M M ZAZ E rep = static value (1.19) A=1 A
Thus the Hamiltonian, equation (1.15), simplifies to N N M NN l, =  V  Z + Enuc rep (1.20) i=1 i=1 A=1I i=1 ij the electronic Hamiltonian, and equation (1.14) is rewritten as
Heie = EelIe (1.21)
and called "the electronic Schrodinger equation".
Even with this considerable simplification of the physical picture the resulting eigenvalue problem, equation (1.21), can only be solved exactly for the special case of one electron moving in the field of the fixed nuclei. Ultimately we would like to address molecular systems which contain hundreds of electrons but for now we would settle for systems containing two to a few. This leads to the next great simplifying approximation. The Independent Particle Approximation
We would like to solve equation (1.21) exactly for the manyelectron case but the electronelectron interactions prevent us from being able to separate the variables, (i.e. the
14
electronic coordinates) and no closed form solution can be obtained. However, equation (1.21) can be solved for the oneelectron case which leads us to a crude approximation. We ignore the electronelectron interactions and solve equation (1.21) as N oneelectron problems, where N is the number of electrons in the molecular system. The total Hamiltonian for the Nelectron system within this independent particle approximation would become
H= hl + h2 + hN (1.22) with
hi = V + A ZA(1.23) A riA(1.23) a sum of oneelectron Hamiltonians, equation (1.22). This form of the Hamiltonian leads to a wave function of the form ' = xl(r1)x2(r2) ... XN(rN) (1.24) a product of oneelectron functions, orbitals, and the total energy of the system will be the sum of the orbital energies.
E = e + E +  EN (1.25) where the {X} in equation (1.24) are hydrogenlike wave functions or an approximation to them.
15
Assuming that the oneelectron Hamiltonian, equation (1.22), and the wave function, equation (1.24), are correct we can now reintroduce the electronelectron interactions as corrections. This is done in an average way, f~ ri d i f) 1 zi x )(1.26) (ijlij) = dr, dr x(i x x(j) X() X(j)(1.26) which represents the repulsion between two electronic charge distributions, Iax(i) 2 and I Xj (j)12. Adding these averaged electronelectron interactions back into the oneelectron Hamiltonian yields the twoelectron Hamiltonian.
S= i + (1.27) 1<3
Hartree proposed the SelfConsistent Field (SCF) method in 1928 [19]. He assumed the repulsion experienced by a particular electron is a spherically symmetric average of the potential due to the N  1 other electrons. This leads to the replacement of the exact twoelectron potential of equation (1.27) with an effective oneelectron potential V = 1Vef = i(i) (1.28) where Jj(i), the Coulomb operator, is defined as J,(i) = dr, x(ij) x() (1.29) When Jj operates on another charge distribution it yields the classical interaction between two charge distributions. It is now possible to express the total Nelectron Hamiltonian
16
as a sum of effective oneelectron Hamiltonians of the form Hef = hi + Vieff (1.30) leading to the set of pseudoeigenvalue equations which determine the oneelectron wave functions {fi}.
(hi + veff)i(i) = eii(i) (1.31) Hartree then assumed a set of wave functions, orbitals, and proceeded to calculate the Hartree operator for the ith electron by evaluating the integrals of the effective potential. The resulting N equations are solved for N new orbitals which are then used to compute the Hartree operator for the next iteration. This process is considered finished when the i + 1 iteration is within a given tolerance of the ith iteration at which time the potential is termed selfconsistent. Hartree's SCF method failed [27, 28], this was due primarily to the form of the wave function.
It is a fundamental principle of quantum mechanics that the measurement of any physical property which depends on the coordinates of a system of identical particles should not depend in any way on the arbitrary labeling of the particles. For two electrons described by the wave function T( Xj(l) Xj(2) ) there exists a probability density I1( x,(1) Xj(2) )12 (1.32) which cannot depend upon the permutation of the electron labels.
j1( X,(1) Xj(2) )12 = IT( Xi(2) Xi(l) )12 (1.33)
17
Which leads to two possibilities
( X,(1) X(2) X,( ) = ( x(2) X(l)) (1.34) or
T( X,(l) x X,( ) = I( x(2) Xj(l) (1.35) the symmetric, equation (1.34), and antisymmetric, equation (1.35), cases respectively. Experimental evidence [29] indicates it is the antisymmetric5 case to which the wave function for two electrons must conform. Hartree's simple product form for the wave function is symmetric with respect to coordinate interchange and this is the principal reason for the failure of the Hartree method.
Working independently of one another both Fock [21] and Slater [22] applied the variation principle to Hartree's method. By varying the radial functions of the simple Hartree product to minimize the total energy they rederived Hartree's original intuitive method on a more rigorous footing, and assured the total energy so obtained is an upper bound to the exact ground state energy.
When the additional constraints of maintaining orbital orthonormality during the variation, along with that of an antisymmetric wave function are applied to Hartree's method, the HartreeFock equations are obtained. The constrained minimization is 5. It is a postulate of quantum mechanics that the wave function describing systems of indistinguishable particles, which themselves possess half integer spin, must be antisymmetric with respect to interchange of particle labels.
18
usually carried out using Lagrange's method of undetermined multipliers. While the antisymmetric wave function is constructed by the application of A on the simple Hartree product. The operator A is the antisymmetrizer.
 (1)p P (1.36)
p
The summation runs over all N! possible permutations of the N particles, p is the parity of the permutation and is equal to the number of twofold permutations needed to bring the orbital product back to normal order. It was Slater who first noted that equation (1.36) was the prescription for the expansion of a determinant [30]. This leads to wave functions constructed from antisymmetrized products of orbitals being commonly referred to as Slater determinants.
The HartreeFock equations are given as
N
f(1) Oa (1)= Zeab bb(1) (1.37) b=1
where f is the Fock operator
f(1) Oa(1) = h(1) + E (Jb(1)  Kb(1)) O.(1) (1.38) b=1
The operator J of equation (1.38) is the same Coulomb operator as that of Hartree's method. The operator K is the exchange operator and gets its name from the exchange of electronic coordinates that takes place during its operation. The exchange interaction in the HartreeFock equations arises due to the antisymmetric nature of the electronic wave
19
function, and so did not occur in the Hartree method. The Coulomb operator was given previously as equation (1.29). The exchange operator is defined as Kb(1) ,(1) = [J dr,2 (22) )]b(1) (1.39) Equation (1.37) can be brought into the form of a standard eigenvalue problem by a unitary transform of the matrix of Lagrangian multipliers.
f(1) Oa(l) = eaa(1) (1.40) The total electronic energy as obtained from the HartreeFock method is given by N NN
E = (To lHIo) = 1(iZhlj) + i[ jij}  (ijji) ] (1.41) where I Io) is the Slater determinant formed from the N occupied HartreeFock orbitals.
Originally the HartreeFock method was applied to atoms and the resulting differential equations were solved numerically. For molecular systems this is impractical for all but diatomics. The difficultly arises in the calculation of the HartreeFock orbitals or Molecular Orbitals. Roothaan [23] and Hall [24] both suggested that the HartreeFock orbitals be expanded as a linear combination of some known set of basis functions according to
ï¿½' = x,XCi, (1.42) where X, is a fixed atomic orbital (AO) basis function and Cvi is the expansion coefficient for the vth atomic orbital in the ith molecular orbital. This, Linear Combination
20
of Atomic Orbitals  Molecular Orbital (LCAOMO), linear expansion reduces the numerical solution of the HartreeFock equations to an algebraic one. The HartreeFock problem becomes a system of linear equations which are solved for the expansion coefficients C,i. The LCAOMO expansion has the added advantage of providing molecular orbitals with a definite analytic form, a huge advantage over numerical orbitals when calculating properties.
Substituting equation (1.42) into equation (1.40) yields
f X.C, = XCie, (1.43) left multiplying (1.43) by X* and integrating over all electronic coordinates yields SF.,,CVi = S.,CieS (1.44) where
F,= J drx*(1) fx(1) (1.45) S = d'rx*(1) x(1) (1.46) If there are N atomic basis functions then there will be N molecular orbitals and N equations of the form (1.43). In matrix notation equation (1.44) is written as FC = SCE (1.47) and is often referred to as the RoothaanHall equation. This is the equation we solve in an iterative manner until selfconsistency is achieved. Upon convergence the columns of
21
C are the coefficients for the AO expansion of equation (1.42), one column or MO per original AO, and the eigenvalues E are the HartreeFock MO energies.
At this point we need to express the closedshell spin Restricted HartreeFock (RHF) equations in their fine detail. This will allow for a better understanding of the ZDO based approximations to them. Also, up to now we have been cheating a little. That is, we have not addressed electron spin.
In 1925 Uhlenbeck and Goudsmit [31, 32] proposed the electron has an intrinsic angular momentum, or "spin", to explain the results of the Stern and Gerlach experiment [33, 34]. Therefore our description of the electron using three spatial coordinates is incomplete, a forth coordinate, the spin,6 is needed. It was postulated that the magnitude of this spin was 1 unit leading to eigenvalues for Sz equaling ï¿½. It is customary to designate this spin coordinate w and to introduce two orthogonal spin functions a(w) and f(w) such that
(aQICa) = ($}) = 1 (1.48)
(1/) = (Pla) = 0 (1.49) Sla) = S(S + 1)1 ) = 'ja) (1.50) SfIl) = S(S + 1)I3) = I1P) (1.51) 6. Dirac placed the intrinsic electron spin on a solid physical footing with his relativistic treatment of the electron in 1928 [4].
22
SzIa) = a) (1.52) z0) =  .) (1.53) If we multiply the set of molecular orbitals {f }, or the set of atomic orbitals {x}, by these spin functions two sets of "spin" orbitals are obtained. The properties of the spin orbitals are such that
i = aila) (1.54) i = il) (1.55) (Obl0i) = (Oil[i)(ala) = 1 (1.56) ( ii~) = ( ~i))($)= 1 (1.57) (O il i) = (Oi l i) = 0 (1.58) or two sets of spin orthogonal orbitals. How the inclusion of spin into the wave function affects its construction will be explored in detail in chapters two and three, equations (1.54) through (1.58) above are sufficient detail for now. The introduction of the spin functions a(w) and P(w) into the electronic wave function has the effect of enforcing the Pauli Principle. Two electrons can be described by, or occupy, one spatial orbital only if they are of different spin.
In Restricted HartreeFock theory all occupied molecular orbitals are doubly occupied; there are no half filled, or openshell, orbitals in RHF theory. Hence the term
23
closedshell. If we take our two sets of spin orbitals, { } and { ), there will be A occupied orbitals in each set, where N is the number of electrons. If we expand the spin molecular orbital basis as was done in equation (1.42), this time using the spin atomic orbital basis, again substituting the constructs into equation (1.40), and integrate over the spin coordinates, the restricted closedshall HartreeFock operator is obtained.
N/2
f(l) = h(l) + E [2Ja(1)  Ka(1)] (1.59)
where
J (1) = dr, 2 (2) 1 ï¿½,(2) (1.60)
Ka(1) = dr2 *(2)  #a(2) (1.61) r12
The operator p2 is a twofold permutation operator, operating on the electron labels. Its effect on the simple orbital product Oi(l) Oi(2) is
Ap2 i(l) oi(2) = ï¿½j(2) j(1) (1.62)
P12 exchanges the electronic labels and the permutation operation takes place prior to the integration. The factor of two in equation (1.59) stems from the fact that an electron in one molecular orbital will have two Coulombic repulsions and one exchange interaction with the pair of electrons occupying a second, different, molecular orbital. There is exchange
24
only between electrons of the same spin. The energy of a RHF Slater determinate is E = 2 E(alha) + Z[2(ablab)  (ablba)] a a,b
(1.63)
= 2 haa + [2Jab  KabI a a,b
The ith orbital energy is given as ei = hii + >[2Jia  Kia] (1.64)
a
We are now in a position to expand the closedshell Fock operator matrix elements into the atomic orbital basis. It is from this vantage point that we will be able to best understand the effect of the various ZDO approximations on HartreeFock theory. The matrix elements of the closedshell Fock operator are given by N/2
(tlfv) = (,lhlv) + Z[2(lJav)  (ptIKalv)] (1.65) where the Coulomb matrix elements for the ath molecular orbital are
(JIV) = (i(1) adr, r(2)  a(2)lv(1)) = (jalva) (1.66) and for the exchange operator
(PIKaI) = (p(1) Jdr, *(2) r2a(1)lv(2)) = (plaav) (1.67) where in equation (1.67) A2 has already operated. The notation is as follows: y and v are atomic basis functions while a is a molecular orbital. Replacing Ja and Ka in
25
equation (1.65) with equations (1.66) and (1.67) respectively gives N/2
fA, = hyV + [2(palva)  (Palav)] (1.68)
a
an expression containing both atomic and molecular orbitals. Expanding Oa, in equation (1.68) above, in the manner of equation (1.42) yields N/2
fl. = h,4V + E E CAaC*a[2(loILA)  (yorIAv)I a oKA
N/2
 = + 2 _ CaC,[(IplvA)  7(p jAv)] a + E 2 (1.69) = hw + Z P,[(yorjvA)  !(yojAv)
= h , + G,,
where the PA are the first order density matrix elements.
N/2
PA = 2 CaCa (1.70)
a
We have now arrived at the point where we can discuss what the ZDO approximation is and which of its manifestations we shall use.
The INDO Approximation
The Zero Differential Overlap approximation is expressed as
XX,dr = 0, ys j v (1.71)
26
In words, for Xp, X, the differential in equation (1.71) is defined to equal zero. Thus the matrix of atomic overlap integrals becomes the unit matrix. The effect of this approximation on the one and twoelectron integrals, in terms of the number of such integrals and the physics they contribute to the theory, is pronounced. All three and four center Coulomb integrals are removed, and no exchange integrals remain, or
S =Av = b, (1.72)
(lAAAclBD) = V6,,,(pA liLA) (1.73) Throughout this introduction we have made reference to the "various" ZDO methods, we are now at a point where we can at least list them, and outline the one pertinent to this work.
Today there are numerous ZDO methods in use. They are all in one way or another based on of the following approximations, the Complete Neglect of Differential Overlap (CNDO) approximation [35], the Intermediate Neglect Differential Overlap (INDO) approximation [36], and the Neglect of Diatomic Differential Overlap (NDDO) approximation [35]. The CNDO method is true to the ZDO approximation, while the INDO and NDDO methods relax somewhat the ZDO constraint on the twoelectron atomic integrals. All SCF calculations performed within this thesis were carried out at the INDO level of approximation and so it is the INDO method we will outline below.
27
Within the INDO approximation the RoothaanHall equation, relation (1.47), is constructed and iteratively solved as was previously outlined. The difference between INDO and abinitio HartreeFock lies in the type of integral included, and their functional form, in the construction of the Fock matrix, with two caveats. The first being that the INDO method described here explicitly uses only the valence atomic orbitals, and the second being that in the LCAOMO expansion only one basis function is used for each atomic orbital. Thus we will be describing, and working with, a valence orbital minimum basis set INDO method. Below the INDO equations analogous to equation (1.69) are presented, itp, v, o and A again label the atomic basis functions, {X}, while A, B, C, ..... label the atomic center the AO basis functions are associated with. The imposition of the INDO approximation on equation (1.69) leads to different forms for specific cases, these are listed below
1. Case p = v; A = B:
FA=A =+  [PcC TAc  VAC] COA
(1.74)
+ PAA [(AAAI AcArA)  '(I(AAAA19A/'A) AA aA
2. Case p Z v; A = B:
FAAA,, A= > PAAA [(AA IVAaA)  I(/1 AAAIO'AVA)] (1.75) )XA bA
28
3. Case L / v; A$ B: FAVB AB  VB AB (1.76) Several new terms were used in expressing equations (1.74) through (1.76), these are the terms used throughout the literature. We shall explain them below starting with the onecenter core integral UPAPA of equation (1.74), UPAA = (AI  V + VAI/1A) (1.77) the onecenter electronnuclear attraction, VA ZA (1.78)
riA
the first order atomic density PC, is defined as, PCe = 1 PAcAc (1.79) Ac
the twocenter twoelectron YAc is defined as, 7AC = (YAAcISAAc) (1.80) and VAc the twocenter electronnuclear attraction as VAC = (PAI VcIPA) (1.81)
29
The fPA term is explained as replacing
3PAVB = (YAl  1V2  VA  VBIVB) (1.82) A book could be written, and many have been [16, 3739], on the subject of obtaining the various terms above. The details of this subject lie outside the scope of this thesis. It should be said the various integrals are replaced by functions of experimentally obtained parameters such as ionization potentials and electron affinities, with the exception of 0 in equation (1.82). The 0 term is, in many implementations, a totally empirical parameter and is usually obtained by varying it, while calculating a specific property or reproducing results of a higher level of theory, until acceptable results are obtained. Such is the case in this work, we shall obtain the SCF orbitals via two INDO parameterizations, one having been parameterized to reproduce molecular structures and the second to reproduce the low energy ultraviolet and visible absorption spectra of molecular systems. Both parameterization follow the philosophy of Pople's original INDO/1 [36] method, at least with respect to the functional form of the U and 7 integrals and in the sense of varying # to reproduce a benchmark value.
We are now in a position to detail how one actually evaluates the spin density, (1.13). At this point we will limit the discussion to a single Slater determinant. The point was made earlier that the unpaired spin density at any point in space (r) was defined to be the alpha density at point (r) minus the beta density at point (r). In terms of the spin
30
molecular orbitals, equations (1.54) and (1.55), the alpha and beta densities evaluated at (r) are expressed as No
pa(r) = 1'(r)12 (1.83) and
NO
p (r) = 5 I(r)l2 (1.84) Then pZ(r) = p0(r)  p^(r) and if we expand equations (1.83) and (1.84) in the manner of equation (1.42) we obtain p'(r) = XPx,(r) x (r) (1.85) and
pP(r) = Pf,x,(r) x*(r) (1.86) where P, and P, are again elements of the first order density matrices of the alpha and beta spaces respectively, or N"
Pf" = ( C ia(Ca)* (1.87)
a
NO
Pu = IC a(Ca)* (1.88) Again the C,, are the expansion coefficients obtained from the solution of the RoothaanHall equation, relation (1.47). The expression for pZ(r) can now be written as pz(r) =(P .  Pf) x,(r) x(r) (1.89)
/A V
31
where all quantities in equation (1.89) above have been previously defined. We should note that for the closedshell RHF theory we previously outlined, the term (Pg'v  PdV) is everywhere identically zero, a consequence of each spatial orbital being doubly occupied.
We can now proceed with the thrust of this work, that is, to develop a method capable of quantitatively describing the unpaired spin distribution throughout a molecular system as measured through the calculation of the isotropic hyperfine interaction, ais,, and to be able to apply this method to systems that are of interest to chemists today. In particular, we will eventually be interested in very large systems, and so our aim will be more toward applying these methods to semiempirical quantum chemistry. However, the theory we develop should, in general, be appropriate for any wave function.
CHAPTER 2
THE UHF APPROACH
Introduction
In Chapter 1 we developed RHF theory in sufficient detail as to allow for the formal definition of pZ(r). Unfortunately, RHF theory, by construction, cannot address molecular systems which possess unpaired electrons. The simplest method for addressing such systems at the SCF level of theory is the Unrestricted HartreeFock (UHF) method which is also known as the Different Orbitals for Different Spins (DODS) [4045] approach. Historically this method dates to the late 1920s where both Hylleraas [46] and Eckart [47] applied it to the He atom, and the method as it is formulated today dates to the early 1950s. Within UHF theory electrons are still assigned to orbitals in Slater determinants [36], but the spatial orbitals associated with aspin electrons are allowed to differ from those describing the #spin electrons. This additional freedom has several consequences; due to the increased number of variational parameters the UHF openshell energy will be < the ROHF1 energy, as well as allowing for a more accurate treatment of electron correlation at the SCF level [4850]. Perhaps the most important ramification, at least for this work, is that a single Slater determinant constructed from UHF spin molecular orbitals 1. Both the Restricted OpenShell HartreeFock (ROHF) method and the concept of electron correlation will be treated in Chapter 3.
32
33
will not in general be an eigenfunction of S2 [51, 52], though it will be an eigenfunction of Sz. As we shall see later in this chapter this so called "spin contamination" will prove to be a problem in the calculation of ais, in the larger radical, molecular or ionic, systems. We now need to develop UHF theory in sufficient detail to understand how Pz,(r) is determined.
The UHF Equations
In Chapter 1 we found the general HartreeFock eigenvalue equation to be
f(1) 4a(1) = eatb(1) (2.1) where the { ,i} are the restricted spin molecular orbitals. We shall use equation (2.1) as our starting point in developing UHF theory. In a manner analogous to that used in defining equations (1.54) and (1.55), the restricted spinorbital definition, we define two sets of unrestricted spinorbitals.
'i  O 1a) (2.2)
i = #p ) (2.3) where the restriction that 4, = # imposed in RHF or ROHF theory has been dropped, thus giving the method its name, Unrestricted HartreeFock. If we now replace the {(;i} in equation (2.1) with the forms defined above, left multiply by (aI and (i#, respectively,
34
and integrate the resulting equations over the limits of the spin variable w we obtain fP(1) 4(1) = e7 01(1) (2.4) and
Pf(1) ï¿½f(1) = ef 0{(1) (2.5) where
f"(1) = h(1)+ [J(1)  K(1)] + J b(1) (2.6) a b
q p
fP(1) = h(1) + [J~() K '(1)] + J(1) (2.7) b a
the aspin and #spin Fock operators respectively. The summation upper limits p and q are respectively the number of aspin and /spin electrons, and by convention p >_ q. Since the kinetic energy of the electron and coulombic nuclear attraction are independent of spin, h(1) is identical to that defined in equation (1.23). The unrestricted coulomb and exchange operators are defined below, the notation is the same as that used in defining equations (1.60) and (1.61).
Jdr(1) = Jdr2 a*(2) r (2) (2.8) 'A(1) = dr2 0*(2) 1 0(2) (2.9) K; (1) = dr2 *(2) (2) (2.10) KP(1) = dr2 ï¿½ *(2) P 0'(2) (2.11)
35
The energy of a UHF Slater determinant is given by P q P
EUHF = + h bb + b+ (Ja;  Ka) a b a,b
(2.12)
1q '9 K") p q a,b a b
where equation (2.12) is analogous to that given for the RHF energy of a Slater determinant, equation (1.63). The Ja term is defined as
// 11 J s = <ï¿½a[ +:) = (IJ2I) = dr,dr, q(1)ï¿½(2) 1_ï¿½(1)q(2) (2.13)
the coulombic repulsion between the aspin and Pspin electrons whose spatial distributions are described by 10,2 and 1ï¿½ 12 respectively, all other terms in equation (2.12) are analogous to their RHF counterparts.
We now expand our two sets of spin molecular orbitals as a linear combination of some known set of basis functions (i.e. The LCAOMO approximation.) consisting of m linearly independent functions {x}.
m
ï¿½7 = E xvc; (2.14)
1 = XvCf; (2.15) v=1
If we now substitute equation (2.14) into equation (2.4) we obtain
SC;i f(1) Xu(1) = Ci Xv(1) (2.16)
A, ii
36
now multiplying equation (2.16) by X*(1) and integrating over r, yields > ; F CY C, =S? SIAC (2.17)
V V
where all quantities in equation (2.17) were defined for equation (1.44), and F;, = dr x (1) f' x,(1) (2.18) In matrix notation equation (2.17) is written as F*Co = SC"'" (2.19) Similar results are obtained when one starts by substituting equation (2.15) into equation (2.5) and carrying out the subsequent integration, or F"C' = SC E" (2.20) These two equations, (2.19) and (2.20), are the unrestricted generalizations of the RoothaanHall equation, (1.47). They were first given by Pople and Nesbet [40], and are often referred to as the PopleNesbet equations.
Once the matrices C" and C' have been determined for a molecular system, pZ(r) is determined from the following. If A is a rectangular matrix of order m x p constructed from the first p singly occupied orbitals, columns, of C", and B is a rectangular matrix of order m x q constructed from the first q singly occupied orbitals, columns, of C, then P = AAt (2.21)
37
PO = BBt (2.22) the adensity and #3density matrices respectively, which are used to construct the total density matrix
PT = Po + Pp (2.23) which follows directly from equations (1.87  1.89) and ai,, is obtained from equation (1.10).
Spin Contamination
While the UHF method has the important advantage of directly producing P, and PO, at the SCF level, it also has several drawbacks as far as this work is concerned. It suffers from the same problem any single determinant method has in describing an openshell system which is not of the highest multiplicity. (e.g. A singlet system consisting of two openshell orbitals describing two electrons.) This is not an insurmountable problem as the molecular systems in which we are interested rarely possess sufficiently high symmetry for there to be nonaccidental orbital degeneracies; of course, accidental degeneracies or near degeneracies will be a problem when they occur. The crucial problem lies in how ,,,, not in general being an eigenfunction of S2, affects pUHF(r). As we shall see later the affect on pF,,(r) can be sufficiently pronounced so as to render the calculated ais, both quantitatively and qualitatively useless. For now we need to
38
present an overview on the origins of spin contamination so the sections dealing with minimizing its effects upon, or removing it from, p',F(r) will be more lucid.
A single determinant UHF wavefunction is of the form
UHF,= .(1)a(1) .....  (p)a(p) (p + 1)(p (p+ 1) ..... (n)/(n)j (2.24)
where in the basis of UHF canonical spin molecular orbitals
(W1) = (4b') = i (2.25) and in general
(#l7 ) $ 0 (2.26) The functions {"} and {'} can be transformed among themselves with unitary transforms
78 = ï¿½U (2.27) 77je= (2.28) and still leave (2.24) unaltered.2
Now by applying
S _ = _S, + Si + Sz (2.29)
2. See Appendix A.
39
to (2.24), it is readily shown that
(%H,,FIS2IIUHF) = S + Sz + q  I( ) (2.30) i j
Equation (2.30) is general and holds for any single determinant wavefunction. (e.g. RHF, ROHF as well as UHF.) It is immediately clear that for RHF and ROHF wavefunctions the last two terms on the righthandside (rhs) of equation (2.30) cancel, (i.e. The overlap matrix in the basis of restricted HartreeFock molecular orbitals is the unit matrix.) and
(S2),F/RIO, = Sz(Sz + 1) (2.31) However for the UHF wavefunction
(S2),U, 2 Sz(Sz + 1) (2.32) where the equality in equation (2.32) occurs when the UHF wavefunction converges to an exact eigenfunction of S2, the RHF/ROHF solution, else
(S2)uHF > (2)EXACT (2.33) Other Interesting Orbitals
The unrestricted molecular orbitals are, of course, not the only interesting type of orbital. We shall find two other types of orbitals to be of use in understanding spin contamination in a UHF wavefunction and in removing it. These are known as the "natural" orbitals [43], which are those orbitals that diagonalize the reduced firstorder
40
density matrix; and the "corresponding" orbitals of Amos and Hall [43] which are also known as the "paired" orbitals of Lowdin [53]. These are the orbitals which diagonalize the overlap matrix between the aospatial orbitals and the #spatial orbitals.
For the wavefunction (2.24) the firstorder density matrix is
p q
7(101') = # * *(1) ;(') a(l)a(1') + *(1) 00(1') 3*(1)3(1') (2.34)
r 8
This is already in diagonal form and, since the canonical spinorbitals are already orthonormal, they are the natural spinorbitals. The natural spinorbitals are not unique since the unitary transforms (2.27) and (2.28) leave the form of (2.24), and (2.35), unchanged.3
The reduced firstorder density matrix is obtained by integrating (2.34) over the full range of the spin functions a(w) and /(w), this yields P q
p(11') = O r**(1)Or.(1') + # *(1)(1') (2.35)
r 8
which would be in diagonal form except that { o'} and {q#} are not mutually orthogonal. The unitary transforms (2.27) and (2.28) can be used to define new orbitals having the property that their spatial overlap matrix is diagonal,4 or Tr= Jdr "77*7 (2.36)
3. See Appendix A.
4. When the unitary matrices U and V, of equations (2.27) and (2.28), are determined as in Appendix C.
41
Tr = Trrs, (2.37) these are the "corresponding" orbitals of Amos and Hall.
In terms of the corresponding orbitals (2.35) becomes p q
P(11') = ;*(1) r a(1') + E 7*(1) 78(1') (2.38) r 8
where the {rr} are spatially orthogonal to each other and to the {r} except when r = s. If these sets of paired functions are replaced with the Schmidt orthogonalized [54] functions
1
Ar = (71 + r4)(2 + 2Tr) (2.39)
1
ir = (  r~7)(2  2Tr)2 (2.40) then all the functions are orthonormal and (2.38) becomes
q q P
P(11') = E (1 + Tr)A*(1)Ar(1') + E (1  Ts) p*(1)p(l')+ ()(')
r s t=q+l
(2.41)
the functions A., ys and 4 of relation (2.41), are the natural orbitals. The corresponding orbitals are then the natural spinorbitals most easily related to the natural orbitals, and the overlap integrals of the "paired" orbitals determine the natural orbital occupation numbers.
A second approach can be taken in determining the UHF natural orbitals, one that will give further insight. First we express the reduced firstorder density matrix in terms
42
of the expansion basis {X}
p(ll1') = (P's + Prs) Xr(1)Xs(1') (2.42) r,s
where Pr, and Pres are matrix elements of relations (2.21) and (2.22) respectively. To see the relationship between definitions (2.41) and (2.42) we first define the matrix C = [AB]mx, (2.43) where [AB] is the concatenated matrix formed by matrix A of relation (2.21) and matrix B of relation (2.22), and n = p + q. Then from equations (2.21) and (2.22) it follows that CCt = (P + P)mxm (2.44) and from Appendix B this has the same nonzero eigenvalues as CtC. Now CC ( I S (2.45) ctc = St I n
where I, and I, are the unit matrices of order p x p and q x q respectively, and S is the overlap matrix between the UHF aspatial orbitals and #spatial orbitals. When the unitary matrices U and V are determined as in Appendix C, relation (2.45) can be brought into the form
(Ut 0 I, S U 0 I, T U t Vt)(I I VS)=UT 0 I) (2.46) where T is the diagonal matrix of corresponding orbitals overlaps. There are then 2q eigenvalues of the form
1 ï¿½ T, (2.47)
43
and (p  q) eigenvalues that are exactly unity, as in (2.41). If the number of expansion basis functions {X} is less than the number of electrons in the molecular system, (i.e. m < n), then there will be m  n overlap integrals, equation (2.36), that equal unity. This corresponds to a set of orbitals whose occupation numbers equals 2. There will be n  m = D of these doubly occupied orbitals, and they have no pairing properties. If m > n, there will be a set of orbitals which have occupation numbers of zero. There will be m  n = E such orbitals, and they also have no pairing properties. When m = n, then all overlap integrals between the paired orbitals will be less than unity and greater than zero.5 i.e.
n>m, D=nm
(2.48)
E=0
n
(2.49)
E = m n
n = m, D = E = 0O (2.50) These relationships can be stated clearly as the eigenfunction equation
O0k = dkOk (2.51)
5. See Appendix C.
44
where " is defined as
n
'= dk Ok(1) 90(1') (2.52) k=1
then
Ok=r= , dk =2= 1+ Tk, k= 1,2,  D (2.53) Ok = Ak, dk =1+ Tk, k=D+l,  q (2.54) Ok = Yk, dk = 1  Tk, k=D+1, q (2.55) Ok =r 7= 00, dk =, k =q + ,. p (2.56) There are no relations for the unoccupied orbitals {E}. Spin Contamination and the UHF PE Surface
For RHF and ROHF wavefunctions the natural orbitals are the canonical molecular orbitals and their occupation numbers are either 2, 1 or 0. It then follows that for an exact spin eigenfunction
OCC
Su = 4q + (p  q) (2.57) where the ai are the natural orbital occupation numbers [55]. For a single determinant no ai other than the RHF and ROHF values can satisfy (2.57) while conserving the total number of electrons. For a UHF wavefunction the natural orbital occupation numbers
45
are functions of the corresponding orbital overlaps, or, ai = l + Ti, i =1,2,  . q
ai = 1  Ti,, i' = 1,2, .. q (2.58) a, = 1, v=2q+1,2q+2," p where i and i' are paired. Thus for a single determinant UHF wavefunction to be an exact spin eigenfunction the Ti must equal 1, i.e. the corresponding orbital overlap matrix must be the same as that of an RHF or an ROHF wavefunction. This implies that the UHF wavefunction must be related to the RHF or the ROHF wavefunction through suitable unitary transformations.
To this point we have expressed the Fock operator as a function of the Fock molecular orbitals. We could have just as easily expressed it as a function of the total density pT, let f(PT) = h + J(PT)  K(PT) (2.59)
(PT) = (P + P) dr2 p*(2) rl2 v(2) (2.60)
(pT) = + P) Jdr2 *(2) v(2) (2.61) Then in the basis of the corresponding orbitals we can write the UHF equations as
(rl7 If(pT) _ K(P)17) = 0 , i = 1,2, p, = p + l,p + 2,. m (2.62)
46
(r7f (pT)+ (Ps)r 0,)= , i= 1,2,q.. q, v= q+ l1,q+2,... m (2.63) where Ps = P"  PO. For an ROHF like solution the {/} and {770} must be the same, or, (2.63) can be written as
( ' If(T)+ l (Ps)I)= 0, i = 1,2,. p, v=p+l,p+2,.r m (2.64)
Comparing equations (2.62) and (2.64), it is obvious that they both cannot be simultaneously correct without Ps = 0, identically. This corresponds to an RHF solution, but from relation (2.57) an openshell spinexact single determinant wavefunction must be related to an ROHF wavefunction. Thus no single determinant spinexact wavefunction is stationary on the UHF potential energy surface. This result will have some bearing on the next two sections.
Spin Projected UHF
As noted above the single determinant wavefunction (2.24) is an eigenfunction of S with (S) = (p  q). However it suffers from spin contamination, described previously, and for the aforementioned reasons is not an eigenfunction of S2. Since we use the PUHF method to "improve" our UHF wavefunction, we will describe below this method (spin projection) for removing the unwanted spin contamination intrinsic to the openshell UHF wavefunction. As we shall see when computed properties are ultimately the goal, it is far easier to work with the densities than it is to work with the wavefunction. The
47
method we detail below is then centered on how the densities are affected as the spin contamination is diminished or removed. It was not our intention to rederive the algebra for projecting out the S = m, component of the spin density matrix. We instead give an overview and list the pertinent equations and citations.
When expressed in terms of the corresponding orbitals the UHF wavefunction can be written in a form similar to that of a Configuration Interaction wavefunction [56, 57], e.g. equation (3.15), or
2(P+q)
XUHF CO' + C bb Cis ... =C (2.65) s=(pq)
where the C, are functions of the corresponding orbital overlap integrals. The T are linear combinations of Slater determinants constructed from the corresponding orbitals, and are eigenfunctions of I2. The excited terms, i, V ..., will contain the Sz =  q) component of the pure spin states SI = (p  q), (p  q +1), (p  q + el), where el is the excitation level. It is these higher multiplicity states, ISI > l(p  q), that will be removed upon projection. Relations for the C, were given by Sasaki and Ohno [58, 59] and are reproduced here since they were used in analyzing UHF wavefunctions in this work. In equation (2.66) below, w. is the "weight" of the components with multiplicity (2s + 1), or, the fraction of the wavefunction composed from the (2s + 1) components,
Ws = C' = (2s + 1)(S )! (l)jDjBsm+j (2.66) (S M')1j=0 2.66
48
where s and mz are the quantum numbers associated with S2 and Sz. While s is a summation index, m, = 1(p  q).
[(s  mz + j)!]2 (2.67) S j!(2s + j + 1)!
() k
Bk = E H{(1  T/)} (2.68) i=1
and the Ti are those of relation (2.36). If the wavefunction is normalized, then
Zws =1 (2.69) To obtain an eigenfunction of S' a projector [41, 60, 61], 0s, is applied to (2.24) or (2.65), where
os = Afl (2.70) tZs
and
(2.71)
S' Os = s(s + 1) 0, A' is the normalized annihilation operator [41, 62]
2 t(t  1)
As = (2.72) s(s 1)  t(t  1) and t, of equations (2.70) and (2.72), ranges over all values, except t / s, consistent with the number of electrons in the system. The projection operator (2.70) projects out
49
the (2s + 1) component of (2.65). The projected state wavefunction can be written as a linear combination of determinants [41].
q
O6s HF = Ck(s,mz, R) T;m' (2.73) k=O
The Ck(s, mz, ) [58, 59] are determined as
Ck(s, mz, ) = (2s + 1) ) (
(s mz)!
(2.74)
[(s  m, + 1)!12
X (1)! (s  m +  k)!(  s  1)! (2s + + 1)! The TI' are a sum of Slater determinants constructed from the corresponding orbitals [63, 64], note that (2.75) is not normalized.
Tm, = 1 ~27,+ p+q I [a Pk/ Ikk] (2.75) The [aPk3pkI is the sum of all spinfunction products, of the p  k a(w) and k P(w) spinfunctions, e.g.
[a''1/3 = aa3 + af3a + paa (2.76) and Iak/3qk] is similarly defined. Then [apk/pkkLPgk] is a sum of
() () (2.77) terms each containing p a(w) factors and q P(w) factors [41, 65].
Harriman [66] showed that the above described projection leaves the eigenfunctions of the spin density matrix, as well as those of the firstorder reduced density matrix, (2.35),
50
unchanged. Only the eigenvalues change upon projection and both sets, projected and unprojected, can be expressed as functions of the corresponding orbital overlap integrals, equation (2.36), and the {Ck}, equation (2.74).
In terms of the natural orbitals of the reduced firstorder density matrix, (or the charge density matrix) relations (2.392.41), the natural orbitals of the spin density matrix are expressed as
b ( = (Ai +/ i) (2.78) z 1 (A= Ai) (2.79) = IV (2.80) In eigenfunction form the relationships are expressed as
0 = k Oz (2.81) where " is defined as
n
S= k Ok(l') 0k*(1) (2.82) k=1
and
0' = 4, (k = (1  Tk2)2, k = D + 1, .. q (2.83) O8 = 4',, k = (1 Tk), k = D + 1, ** q (2.84) O = , = 1, k = q + 1,.. p (2.85)
51
As was the case for the charge density natural orbitals, there are no relations for the unoccupied orbitals {E}. The Tk of the doubly occupied orbitals {D} all equal unity and thus have occupation numbers of zero, see equations (2.83) and (2.84) above, and so do not contribute to the spin density matrix. The validity of relations (2.782.80) can be checked by expanding them in terms of the corresponding orbitals and applying =Ip  p", where
p
P = 7(1') ?r*(1) (2.86)
q
P = Z s(l') *'(1) (2.87) and check against relations (2.832.85).
If we now follow Harriman's [66] line of reasoning we can obtain the above results from a different perspective. We also will arrive at a point where Harriman's observations on the effects of projection are apparent.
We now want to construct the operator matrix 3 in the basis of the charge density natural orbitals, and then obtain its eigenvalues and eigenfunctions. First we should note that fr can couple differing orbitals only if the differing orbitals are one of the pairs of "paired" orbitals, relations (2.39) and (2.40). Then the largest nonzero subblocks will at most be of order 2 x 2, and lie along the principle diagonal. This reduces the problem
52
to diagonalizing a series of 2 x 2 matrices of the form ((AimIAi) (AiIii) ) (2.88)
to obtain the eigenvalues, and since (2.88) is selfadjoint it can be diagonalized via a unitary transform. If we use the Jacobi method [67] to obtain the eigenvectors we will gain an insight on the effects of the projection on these eigenvectors. We can use the unitary matrix
(cos ti sin 9i (2.89) sin 9i  cos it
to diagonalize (2.88), and obtain its eigenvectors and eigenvalues. Once 9i is determined the eigenfunctions are obtained from cos(0i) Ai + sin(t9i) pi (2.90)
sin(t9g) Ai  cos(i) pi and di is determined from t (illm) + (pl~diIA) tan 29di (AII)  (2.91) now
(Ail"7Ai) = (il7"lyi) = 0 (2.92) and
(Ail7 li) = (uil?1Ai) = (1  Ti2). (2.93)
53
then matrix (2.88) is of the form ( 0 1  T)(1 2) (2.94) (1  Ti) 0
which has eigenvalues of
+(1  T/2) (2.95) From relations (2.91) and (2.92) we see that tan 2#i = oo, or Vi = ), for all pairs (i, i'). This leads to
(cos i9 sin ) Ag ) = (l) = (Ai + ) (296)
sin di  cos0 i ] \i] \ï¿½i, = (Ai Ii)
being the eigenfunctions of (2.88), or, the same eigenvalues and eigenfunctions found previously in relations (2.782.79) and (2.832.84).
Harriman solved the matrix elements of (2.91) for the projected spin density operator6 and obtained 0 = .1, the same as for the unprojected density. Thus the {(f} and {q4'} are unchanged by the projection, as are the singly occupied { q,}. (The interested reader is referred to References 62, 66 and 68 for a detailed discussion of spin projection on a UHF wavefunction.)
We give Harriman's relations for the projected spin density matrix eigenvalues below. These are the relations used in this work to explore the effects of spin projection on the 6. Table IV of Reference 66. Harriman solved for the projected eigenvalues, and eigenfunctions, of p o+ and p as well as for p.
54
calculated aiso.
s+1 W1v=l+ 1 1 (2.97) S +1
(1 [1 Ti])1 s+1l s
The relations (2.97) are for the special case of ISI = Sz [661. The Ci and Ci, are paired eigenvalues that in the unprojected state sum to zero, while the ,, have no pairing properties and in the unprojected state are equal to unity. The w,(i) term is defined as q1
ws(i) = E(1)kCk(s,mz,  1) Ak(i) (2.98) k=O
where
(q1) k
Ak(i) = { {TlJ} (2.99) and the Tj are those of equation (2.36). The w, term was defined in relation (2.66).
It was not our intent to derive the relations of equation (2.97), to do so would fill half a Dissertation with unoriginal work. Harriman and others [41, 51, 56, 57, 6062, 66, 6873], and references within, have worked through the tedious algebra and arrived at (2.97). We have implemented (2.97) for the expressed purpose of investigating the effect of spin projection on the aiso calculated for the large molecular systems we are interested in describing.
55
The AUHF and EUHF Methods
There are two methods related to PUHF that we have not mentioned, and, except for the following example, will not pursue. These are Annihilated UHF (AUHF) [74, 75] and Extended UHF (EUHF) [70, 76]. In Annihilated UHF one or more of the higher multiplicity components are removed from the wavefunction through the application of the annihilator A, [41]. As in PUHF the As is/are applied after the SCF, and thus the annihilated wavefunction is no longer stationary on the UHF PE surface. In practice it is usually the 2Sz + 2 component that is removed, and it is hoped that the higher multiplicity components lie too high in energy to contribute significantly to the spin contamination [43]. We have applied the A,+,, s = Sz, to the INDO/1UHF wavefunction describing the Bacteriochlorophylla (Bchla) and Bacteriochlorophyllb (Bchlb) cations [77], with the following results for (S') and the aN of the four pyrrole nitrogens. Table 21: The affects of annihilation on (S) and calculated aise.
Bchla+ Bchlb+ Experiment7 UHF AUHF IExperiment UHF AUHF
(S2) 0.750 3.970 13.36 0.750 4.268 14.79 aN18 0.78 2.24 4.68 0.82 1.80 8.75
aN2 1.12 18.4 38.3 1.17 16.2 78.4
aN, 0.83 12.1 25.4 0.82 10.6 51.6
aN, 1.03 2.73 5.52 1.06 2.43 11.5
7. From Reference 77.
8. The units are Gauss, and the aN are for 14N.
56
From Table (21) we can see the assumption that the higher multiplicities contribute little to the total wavefunction is wrong. As (S2) increased from 3.970 to 13.36 upon removal of the Il = 1 component of the INDO/1UHF wavefunction for Bchla', and from 4.268 to 14.79 for Bchlb+. Further we see that the INDO/1UHF calculated aiso are all too large, and more than double upon annihilation. It is clear that the components with multiplicities greater than (2Sz + 2) are contributing significantly to UHF, and the computed aiso reflect this. (We should note that for small molecular systems annihilation does improve the (S2) as well as the calculated energy [78, 79].) Then from Table (21) AUHF is fatally flawed, at least when viewed from the perspective of wanting a method applicable to large molecular systems.
The Extended UHF method mentioned previously is basically a spinprojected unrestricted function where the orbitals are chosen to minimize the energy after projection. The EUHF wavefunction is stationary with respect to its variational parameters, and the spin densities calculated with it tend to lie between those determined via UHF and PUHF [70]. The EUHF methods are computationally cumbersome, and as stated above the resultant spin densities are no better the those of PUHF, at least in the sense of removing the unwanted spin contamination. For these reasons the EUHF method was rejected as a possible tool for exploring spin distributions within large molecular systems.
57
Calculated Results for the UHF and PUHF Methods.
In Table (D8) we present the results for calculating the aH and aN of several more test structures using the parameters of Appendix D, and the unpaired spin densities produced via the methods discussed in this chapter. In Appendix D we found that the INDO/1 Hamiltonian yielded consistently worse results, compared to those of the INDO/S Hamiltonian.9 For this reason we chose to limit our presentation to the INDO/S results. Along with the additional test structures we also present results for Bchla and Bchlb cations and anions, as well as the Bphea anion. These are the types of compounds we want to be able to describe properly. The structures were obtained from the Brook Haven National Laboratories crystallographic data bank, through a collaboration with Dr. Jack Fajer, their number system can be found in Figure (15). We note that the hydrogens of the three crystal structures were not determined experimentally but were placed in position via a best guess algorithm. Upon inspection of the Bchl structures we found several "strange" methyl groups and repositioned the hydrogens associated with them in a manner consistent with chemical common sense. All calculated results will be presented in units of Gauss (G).
9. See Tables (D3) through (D5).
58
Table 22: Calculated results for Bchla.
Magnetic Calculated a,,, System ( %2) % Xuhf Center'o Position" UHF PUHF a.,p12 Bchla+ 2.11 62 N 21 0.35 0.37 0.78 N 22 6.10 6.02 1.12 N 23 2.88 2.86 0.83
N 24 1.61 1.58 1.03 H 5 5.84 6.82 0.84 H 10 7.30 8.54 0.46 H 132 3.23 2.07 0.58 H 20 1.55 1.74 0.46
Bchla 2.56 55 N 21 2.05 2.03 ï¿½0.42 N 22 1.42 1.43 2.32 N 23 0.38 0.40 ï¿½0.19 N 24 1.76 1.72 2.09 H 5 6.01 7.16 3.43 H 10 7.19 8.42 2.46 H 132 2.98 1.64 2.22 H 20 2.12 2.47 <ï¿½0.18
10. The isotopes are 14N and 'H. 11. See Figure (15).
12. Reference 77.
59
Table 23: Calculated results for Bchlb and Bphea.
Magnetic Calculated ao.. System (ï¿½2) % 2Jh, Center Position13 UHF PUHF a,14 Bchlb+ 2.44 59 N 21 0.74 0.75 0.82 N 22 9.43 9.23 1.17 N 23 2.24 2.24 0.82 N 24 1.51 1.49 1.06 H 5 7.42 8.28 0.89 H 10 8.64 9.59 0.43 H 132 2.25 1.76 0.43 H 20 1.03 1.10 0.43
Bphea 2.47 58 N 21 0.58 0.55 0.43 N 22 7.15 7.07 2.56 N 23 0.78 0.75 0.21 N 24 1.83 1.71 2.19 CH3 2' 3.75 2.71 2.53 H 5 8.82 9.59 3.02 H 8 5.31 4.17 0.65 H 10 8.10 8.97 2.85* CH3 12' 4.67 5.30 2.96* H 20 5.80 6.58 2.47 NH 21' 0.89 0.84 0.35 NH 23' 1.10 1.07 0.17
13. See Figure (15).
14. Reference 77.
60
The results presented in Tables (22) and (23) are somewhat disappointing. Neither method reproduces the experimental observations very well. All four cases were rather badly spin contaminated, as implied by the value of (ï¿½2). While the PUHF method is an exact eigenfunction of S2, the results of this method still paralleled those of the UHF method. The only apparent affect of the projection was to diminish the IpZi by approximately 2, as is expected by the form of equations (2.97). Obviously, after projection, the pZ of the two methods stayed proportional to one another. This is surprising, and disappointing as we had hoped that the projection would yield values in better agreement with experiment, having removed all spin contamination from a highly contaminated UHF wavefunction. Neither the signs of the splittings nor the their magnitudes were properly described by either method. At least not in a consistent manner. We should also note that the * values are considered to be suspect by those who took the measurement.
It is obvious from these results that neither the UHF or the PUHF methods are capable of properly describing the unpaired spin distribution within these molecular systems.
CHAPTER 3
THE ROHFCI APPROACH
Introduction
In this chapter we shall treat the openshell electrons differently at the SCF level than we did in Chapter 2. In Chapter 2 we used Pople's DODS approach and obtained a single determinate wave function which was an eigenfunction of S, as it must be, but not an eigenfunction of S2. In this Chapter we shall use an SCF procedure that produces a wave function that is an eigenfunction of both these operators. Again as in Chapter 2 it will help if we outline the details of this method, the Restricted OpenShell HartreeFock (ROHF) method, as it has been developed and implemented by Zemer and Edwards [80].
Restricted OpenShell HartreeFock
Restricted OpenShell HartreeFock was first proposed by Roothaan in 1960 [81]. Since that time numerous methods have been proposed to address the problem of openshell molecules at the SCF level of theory. A sampling of these methods can be found in References [8192]. We outline the ZernerEdwards method below.
The total wave function is constructed from a linear combination of Slater determinates. The coefficients of the Slater determinates are chosen to represent a particular spin state [93], to guarantee that this is an eigenfunction of S2. The coefficients are 61
62
determined by spin symmetry. Focklike operators are constructed for each shell, i.e. closedshell, 1st openshell, 2nd openshell etc. The constraint (OijF'  F"Ij) = 0 (3.1) is enforced and guarantees the matrix of Lagrange multipliers is Hermitian, where F' is the Focklike operator for the pth shell and F" for the vth shell. The Hermiticity of each Focklike operator guarantees an orthogonal set of molecular orbitals for each operator. Though the various sets of molecular orbitals, one set for each openshell, will not, in general, be orthogonal to one another, as each has been constructed from the diagonalization of its own operator F", or ( 1 ) = (0'i4) = 0 (3.2) but in general
(l) = ( I 0) 0 (3.3) A concatenated set of orthonormal molecular orbitals is guaranteed in this procedure by projecting out the eigenvectors of each successive Focklike operator in the orthogonal complement of all those proceeding it.
The spin density, equation (1.13), determined from an ROHF wave function is everywhere positive, a consequence of the construction of the ROHF wave function. Although negative spin densities are generally small, this prediction is clearly wrong.
63
Direct experimental observation shows the sign of the unpaired spin density can be negative as well as positive. We have lost some of the flexibility we had with the UHF wave function, a price we paid to obtain a simultaneous eigenfunction of the Sz and S2 operators at the SCF level. A single spinadapted configuration state function lacks the flexibility to describe differences in unpaired spin distributions. We would have to construct a total wave function of many such "spinadapted configurations" to have the degree of flexibility necessary for a proper, or realistic description. The simplest way to construct such a total wave function is through the method of Configuration Interaction
(CI).
Configuration Interaction
The Configuration Interaction method is a straight forward application of the Ritz [94] method of linear variations as applied to the determination of electronic wave functions. Again we desire to solve the Schridinger equation, equation (1.14), and again we must approximate its solution. In the CI approach the approximate wave function I,,, is expanded as a linear combination of "configuration state functions" (CSF) in an analogous manner as the LCAOMO approach, equation (1.42), for the HartreeFock molecular orbitals.
N
ps 8= 8 C(3.4)
S
64
Where the {} are the configuration state functions and the expansion coefficients, Cs, are the parameters which are varied to make E[,p,,] stationary. N is the number of expansion functions. We examine, E[ !apx] = (''aPzIHI~apz / ('I'apxj'Pap.) (3.5) in which H in equation (3.5) above is the electronic Hamiltonian, equation (1.20). This leads to the general matrix eigenvalue equation HC = SCe (3.6) where
Hlit = (AsIHIt) (3.7) and
,t = (, ) (3.8) The N linearly independent eigenvectors c,, the columns of C, can be chosen to be orthonormal with respect to S,
N
c S c, = * CS,,ct, = pq, (3.9) s,t
leading to the simple eigenvalue equation HcP = CPEP (3.10)
65
If we label the corresponding eigenvalues such that El < E2 < E3 < . _ EN (3.11) each E, is an upper bound [95, 96] to the corresponding eigenvalue of Her. Furthermore as additional configuration functions are added to the expansion, equation (3.4), each eigenvalue EpN+ of the (N + 1) term expansion satisfies EP_" < EN+I) < EN) (3.12) and as {(N} approaches completeness the {E"} approaches the exact eigenvalues of Hei from above [95].
Correlation Energy
The difference between the "HartreeFock limit" energy, which is the limit approached by restricted selfconsistent field calculations as the basis set approaches completeness, and the exact solution of the electronic Schrotdinger equation has been called the correlation energy [51]. It reflects the fact that the HartreeFock Hamiltonian contains the average, rather than the instantaneous, interelectron potential, and thus neglects the correlation between the motions of the electrons.
It is customary today to break down the correlation energy into two subcategories, dynamical and nondynamical. Dynamical correlation is the concept of correlated electronic motion, while nondynamical correlation effects are due to near degeneracies and
66
rearrangement of electrons within partially filled shells. The CI method recovers the correlation energy, both dynamic and nondynamic, at least in the limit of a full CI. (We will explain the various CI expansions later.) While the inclusion of correlation into an abinitio wave function is necessary for the accurate description of observable properties, it is a less than good idea for semiemperical methods. The reason this is so is straight forward; semiemperical methods are parameterized. These parameters include ionization potentials and electron affinities as well as completely empirical parameters such as 9,, equation (1.82), which are varied so as to reproduce some benchmark. If the parameterization is based upon experimental values then correlation is included to an unknown degree. An example of how correlation is built in can be seen in the MatagaNishimoto formulation [97] of the twocenter twoelectron y,,AC integral, equation (1.80)
RACACAC A 1 (3.13)
2
where
"AA = IPA + EAA (3.14) the sum of the ionization potential and the electron affinity. The experimentally measured IP and EA are already "correlated"; any and all such effects or interactions are already embodied in these measured quantities. To now introduce correlation into the INDOSCF wave function is to, in an indeterminate way, "doublecount" the correlation. The amount
67
of correlation that is built in depends on the level of theory that is parameterized and what parameters are chosen.
Single Excitation Configuration Interaction
We now need to address the configuration interaction method used in this work. The CI basis, {}, is constructed from a set of spin molecular orbitals obtained from a INDOROHFSCF calculation. The expansion functions 4 are linear combinations of Slater determinates whose coefficients are chosen such that the CSF's are eigenfunctions of both Sz and S2, where (Sz) and (S') of {} are the same as that of the INDOHF reference function. Such configuration functions are said to be "spinadapted". The method used to construct these spinadapted functions and the algebra imposed by their form in the evaluation of the spin density will be addressed in detail later. We now need to examine the expanded wave function in more detail. We can write equation (3.4) in a less general and more insightful form as
OCC vir ocC vir oCC vir
c=1 t+ ijk 5ijk + (3.15) i a i,j a,b i,j,k a,b,c
where 40 in the first term of equation (3.15) is the ROHF reference. The set {} of the second term is comprised of all spinadapted configurations generated by removing one electron from an occupied spin molecular orbital Oi and placing it in the virtual, or unoccupied, spin orbital ,ba. These are known collectively as the single excitations. The {b} is generated by removing two electrons from the occupied molecular orbitals, or
68
occupied space, and placing them in the virtual space. Such spinadapted configuration functions are also known as double excitations or doubles. The same pattern holds for the triple and higher excitations. If the summation of the various excitation levels in equation (3.15) is carried to its limit, then the calculation so performed is said to be a full CI calculation. The number of configuration functions N used in the construction of a full CI wave function, for a given number of electrons n, orbitals M and total electronic spin angular momentum S is given by Weyl's formula [98] 2S+1 M+1l( M+1 N
N(n, S, M) = 2S+I M+ (3.16) M+1 \S ! ,(!+S+ 1
which also equals the dimension of the full CI Hamiltonian matrix. Obviously full CI calculations are, from a computational point of view, intractable for all but the smallest of molecular systems. For this reason the CI expansion, equation (3.15), is most often truncated to include at most only the first two or three terms. Even with this reduction in the size of the problem it is not unusual to reduce it even further by limiting the size of the "active space". That is to limit the number of molecular orbitals from which electrons are removed from or placed into, usually centered around the highest energy occupied molecular orbital (HOMO) and the lowest energy unoccupied molecular orbital (LUMO), or the "HOMOLUMO gap". For the work in this thesis the CI expansion, equation (3.15), is truncated at the second term and all single excitations, {f }, are included in the CI wave function, and it is referred to as a "singles excitation CI" or
69
a CIS calculation.
We can now come back to the problem of the "doublecounting" of correlation. If we construct a simple trial wave function comprised of the closedshell SCF reference, Io, and one singly excited configuration function, (D, or qas = (Co4o + Cafo) (3.17) and proceed to evaluate (1 cslH,,Iis) for its lowest eigenvalue and associated eigenvector we obtain the matrix eigenvalue problem ((o o) (oI ) C = Eo (3.18) It is obvious that any mixing of the diagonal elements is through the offdiagonal element (D01H.lI) = (ilhla) + 1 [(irlar)  (irlra)] (3.19)
r
or its adjoint. The iath element of the Fock operator, again in the basis of spin molecular orbitals, is given by
( ifl.) = (illa) + Z[(irlar)  (irlra)] (3.20) Notice that the righthandside of both equation (3.19) and equation (3.20) are identical and thus
(@ol nIH,?) = (1 Ifl'oA) (3.21)
70
Now by definition solving the HartreeFock eigenvalue problem requires the offdiagonal elements of the Fock matrix to vanish, or upon solution ( If 10.) = 0 (3.22) and then so must equation (3.19). The above exercise is an example of Brillouin's Theorem [99]. Which is stated in words as; singly excited Slater determinates IQ) will not directly interact with a closedshell HartreeFock reference determinate Ijo) through the electronic Hamiltonian, or
(olHC,, l ) = 0 (3.23) as well as its adjoint. Restricting the CI expansion to the { } leaves the closedshell HartreeFock reference uncorrelated, and so the problem of doublecounting correlation has been avoided. This approach to the CI method has proven quite successful for the calculation of the low energy ultraviolet, and visible, absorption spectra of molecular systems [100103]. This success was the primary reason we chose this technique as a starting point for the calculation of (p(rN)).
Unfortunately, at least from the point of view of doublecounting correlation, equation (3.23) does not hold for an openshell HartreeFock reference. Certain single excitations from an ROHF reference function lead to spinadapted configuration functions which differ from the reference by two spin molecular orbitals and such configuration functions
71
interact directly with 1o), and in general
(o lne LO) $ 0 (3.24)
An example can be given by letting the wave function equation (3.37) interact with the wave function equation (3.41) through the electronic Hamiltonian, vis
2
(T. IH.01T) 6 (imlma) (3.25)
It is this direct mixing of the reference function with "certain" singly excited functions, and through them indirectly with the other single excitations, that gives the CIS approximation to the molecular ground state the flexibility to describe differences in the alpha and beta spin distributions; unlike the ROHF reference function.
This leads to the question; can a method that was carefully crafted to avoid such interactions as equation (3.25) above now be expected to well describe another property, and can the introduction of correlation to an already "somewhat" correlated reference function be done in such a way as to obtain consistent results? The first question above is answered by this thesis and the second is answered in Table (31) below.
72
Table 31: How the correlation of the INDOROHF reference affects the calculated value of aiso.
% Correlation
System aH Exp1 A aH CIS CISD2
20H 26.4 3.34 8.31 100.0 2CH3 23.0 2.49 7.47 98.4 2CH 20.2 1.98 4.23 98.6 2BH3 15.1 2.05 5.50 98.4 2NH 25.1 3.53 17.7 100.0 2NH3+ 25.9 3.10 9.99 96.0 2HCN 137.1 4.14 3.92 93.5 202H 9.8 1.35 1.38 96.7
In Table (31) the (aH Exp) column lists the experimental Hydrogen aiso values, and (A aH) represents the difference in the Hydrogen's calculated isotropic hyperfine splitting constants when calculated from a CIS and a CISD wavefunction. The (% Correlation) columns lists the fraction of the correlation energy recovered, in percent, when compared to a full CI. It is of interest to note the average (A aH) is 2.75 and its standard deviation is 0.93; recall that a,,so varies almost linearly with respect to pz, equation (1.10). A comparison of columns (aH Exp) and (A aH) indicates that regardless of the magnitude of aso the shift is a rather constant 3 Gauss. Inspection of the CIS and CISD (% Correlation) columns shows that the CIS wave function has recovered a small fraction
1. All ai,, values are reported in units of Gauss.
2. CISD calculations include the first two terms of equation (3.15) in the construction of the approximate wave function.
73
of the correlation energy, which is expected due to the generalized Brillouin theorem [104] while the CISD wave function has recovered almost all of it. Nevertheless, the calculated property aiso is for all intents and purposes the same for either wave function. This demonstrates that the calculated CIS a,,so values increase in magnitude by a small and almost constant amount when the wave function is fully or almost fully correlated. This small "constant" shift indicates that a CIS wave function has the same flexibility to describe the spin distribution as a more highly correlated wave function at least within the INDO/S model. This to a large extent relieves our concern with the possible problem of correlating an already correlated reference, and the possible inconsistencies that may have been introduced into our calculated as,,. We chose Hydrogen for the above endeavor for the following two reasons; 1) there is no core orbital spin polarization problem when calculating aio,, using a valence orbital only theory and, 2) minimum basis set theories have problems describing the "breathing" of the atomic orbitals as a function of the atomic charge. By taking the difference in the computed aiso the magnitude of the contact has to a very large extent been removed, and the difference represents almost purely the difference in the spin density as computed by the two wave functions. This leaves (A aH) a good gauge of the effect of correlation on the calculated aiso, even when using a minimum basis valence orbital only theory.
74
SpinAdapted Configuration Functions
There are many ways to construct simultaneous eigenfunctions3 of Sz and 2 when starting from a set of HartreeFock spin molecular orbitals { 1}. Each method entails a unique procedure for the evaluation of the matrix elements of the CI Hamiltonian, a subject we shall not be concerned with, as well as for any integrals necessary to evaluate the expectation values of observable properties, a subject that shall concern us. In this work two Configuration Interaction procedures were used in the evaluation of aise. One is quite general and in principle can treat molecular systems of any multiplicity and of any openshell structure; it is also quite slow. The second CIS procedure is specific to a singlereference, singlesexcitation, calculation on molecular systems with (Sz) equal to , or doublet molecular systems; it is also quite fast. Since at the time of their implementation the procedures for evaluating (F(rN)) for either CIS method were unique, we shall present them in some detail here.
Rumer Bonded Functions
A generalized extended Rumer diagram methodology [105107] has been presented by Cooper and McWeeny [108, 109]. Their formulation was general to any matrix element between bonded functions [106, 107] and included formulae for spindependent operators. We, while not being cognate of this work, also worked out formulae for matrix 3. An outstanding reference on the subject of "Spin Eigenfunctions" can be found in the book by Pauncz, Reference 93.
75
elements of a spindependent operator, built upon the initial work of Zerner and Manne [110]. Since it was this algebra that was implemented we shall develop it below and not that of Cooper and McWeeny, through as one would expect they are similar.
The valencebond type spin functions grew out of the work of Heitler and Rumer [105107] and of Slater [111], and are commonly called Rumer or Slater bonded functions.4 The configuration functions constructed to be spin eigenfunctions via the Rumer method are linearly independent [93], but not in general orthogonal. The Rumer functions constructed from the same set of spatial orbitals will not be orthogonal among themselves but will be orthogonal to all other Rumer functions, assuming an orthonormal set of spatial orbitals form the initial orbital product. The spin coupling in a particular configuration function based on the orbital product
102 . .. ï¿½ ,, (3.26) is described by a system of parentheses
(0 (42 ( 0 ... #,~,_s),,) (3.27) singlet spin coupling pairs of orbitals in the product. For a given number of electrons N and a total electronic spin angular momentum S, there will be N  S pairs of spin
4. We shall use the notation of Boys, Reeves and Sutcliffe [112, 113].
76
coupled orbitals for the case of S = mr,.5 For i $ j
(0i ,,) = (r)0, (r2) [a(w)3(w2)  /3(wl)a(w2)] (3.28) for i = j
( ,') = A4q,(r),(r2)[a(w)#3(w2)] (3.29) The unpaired orbitals are represented by "dangling" left parentheses, again for i f j
1
b)(k= (r (r (r)[( (2)(w3) l(w1 )a(w2)(w)] (3.30) During orbital rearrangements to obtain maximum orbital coincidence between two spinadapted configuration functions each orbital keeps its associated parentheses.
Diagrammatically the Rumer spin functions are constructed in the following manner; the spatial orbital labels are placed at the vertices of a polygon with one extra vertex added as the "pole". Starting at the pole and proceeding around the polygon in a counter clockwise manner the orbital labels will be in ascending order. The labels are then connected by arrows into singlet coupled pairs, equation (3.28), with the arrow's tail at the lower index and its head at the higher index. As many such couplings as necessary are formed to obtain the desired Sz value. All uncoupled orbitals are connected to the pole by 5. For a given N electrons and S electronic spin angular momentum there will be f(N',S) = 1) spin functions [114] for a given orbital product. Where N' equals the number of singly occupied orbitals and q' equals h  S.
77
arrows. Within the above rules all such possible diagrams are constructed and any with crossed arrows are rejected. Those with crossed arrows are not linearly independent [115] and can be constructed as a linear combination of those diagrams whose arrows are not crossed. For the case of a three electron three orbital S = I system see Figure (2) below. Equation (3.31) translates the diagrams of Figure (2) back into the conventional algebra
1 * 1 * 1
2 3 2 3 2 3
D1 D2 D3
Figure 2: Rumer bond diagrams for the case of three electrons in three orbitals and S = 1
D1 = ( 2 =22 23(a a  #aa)
1
D2 = (0243)( = I,2,4,1I(apa  Oaa) (3.31)
D3 = ( = 22 (aPa  feaa) where the electron labels are assumed to be in canonical order and have been suppressed. By inspection it is obvious the D1 + D2 = D3. Armed with the formalism of this section we are ready to describe the evaluation of (p^(r)) using a Rumer CI wave function.
78
Evaluation of pz for a Rumer CI Wavefunction
The evaluation of matrix elements between Rumer functions for the spin density operator, p, is achieved as follows. Place the Rumer functions in maximum coincidence, then connect each spin molecular orbital in the first function to its counterpart in the second with a vertical bar. vis (4>21)(4,4)(#s = (1 2) (5 (3 4) I I I I I (3.32) (ï¿½22)(ï¿½30)(ï¿½4 I= (1 2) (5 3)(4 Then horizontally connecting the coupled pairs and suppressing the indices leaves a "closed loop" and a "odd chain". vis
Figure 3: A closed loop and an odd chain. For cases where S is greater then "even chains" are also possible. vis
Figure 4: Two types of even chains which arise from (1 2) interacting with (1 (2 in the first case and (1 2) (3 4) interacting with (1 (2 3) (4 in the second. and in general so are multicycle closed loops. vis
79
Figure 5: A multicycle closed loop, arising from the interaction between (1 2) (3 4) and (1 (2 3) 4).
Once the matrix element between a pair of bonded functions has been reduced to the above structures, the evaluation of (P) is straight forward. The operator P will couple one orbital pair, linked by a vertical bar, at a time and (P) is then just the sum of these couplings.
Now from the properties of the closed loops their total contribution to ( ) sums to zero. This is a direct consequence of
(0IJ) = (Oji) (3.33)
If an even chain is formed then the contribution of the matrix element between the two Rumer functions is zero. There will always be spin orthogonal orbitals coupled in such a case, and no oneelectron operator leaving the primitive spin functions unchanged, such as P, can couple even chains. This leaves only the odd chains contributing to (P (rN)). Thus only the density contributions of the orbitals coupled by vertical bars in the odd chains need be computed, and (P(rN)) as evaluated from a Rumer CI wave function
80
is written as
(z) c c cjFI. qS(OlM30q3) (3.34) or in the atomic orbital basis
(Pz(rN)) = CrJFI [ qZ P, (Xp6(r  rN)I ) (3.35)
I,J L,J
where the summation index (s) runs over all the vertically linked orbital pairs, qS is a link parity factor and alternates as +  +    +, the summation indices P and v run over all atomic orbitals, the coefficients C, and C, are the CI expansion coefficients, see equation (3.4), for the Ith and Jth Rumer functions respectively and the indices I and J run over all such bonded functions. The atomic orbital density P, is determined from the sth vertically linked molecular orbital pair. The factor Fr, F,1 = (2)1b(2) +(1)+J (3.36) accounts for the product of the normalization factors of the Ith and Jth bonded functions as well as phase changes due to orbital permutations within the bonded functions necessary to obtain maximum coincidence between them. In equation (3.36) (1) is the number of closed loops, (b) is the number of bonded pairs (O4,j), (k) is the number of bonded pairs with identical orbitals in one function not matched by a bonded pair of identical orbitals in the other, the factors a1 and a, are the number of permutations necessary to bring the unpaired orbitals back to their original order in bonded functions I and J respectively.
81
Fast Doublet CIS
Whereas the Rumer bonded function method for the construction of spinadapted configurations was very general with respect to the multiplicity, number of openshells and degeneracies within the openshells, the following method for their construction is very specific, and entails a new formalism for the evaluation of (P(rN)). For lack of a better name we shall call this method the "Fast Doublet CIS" (FDCIS) method. The FDCIS algorithm is specific to the following case, only singles excitations off a single reference determinate of spin S = . This restriction allows for concise and easily evaluated matrix elements of the CI Hamiltonian [116], again this topic lies outside the scope of this work. We shall detail below the formulae for the evaluation of (P(rN)) from a FDCIS wave function.
FDCIS SpinAdapted Configurations
Within the FDCIS formalism there are four unique molecular orbital configurations, see Figure (6), these lead to five "types" of spinadapted configuration functions. We shall label these generically as To ... T4 and shall define them below. In an attempt to achieve
82
To Ti T2 T3&4 Figure 6: The four possible orbital configurations for the FDCIS formalism. compactness of form the following convention shall be followed; a Slater determinant,
, n 4, will be represented as ii j ... n. The orbital indices i,j, a and m shall take on the following meaning, i and j will represent doubly occupied orbitals, (a) shall represent a virtual orbital and m shall represent the openshell orbital in the reference function. Any subscripts on the orbital indices will indicate the "type" of configuration function from which it came, it will be clear below what this means. Now let us write the five types of spinadapted configuration functions To = 2'o = Ili... mI (3.37) Ti = 2 ,' = 11... m ... mj (3.38) T2= = li11 i ... 1m (3.39) T = . . (11... ia. ... m + Ili ... ai.. mI) (3.40)
83
T4= 2 7 = (211i...ia.... ii .i . m + 11i a... i mI) (3.41)
where the superscripts and subscripts on <4 indicates the single excitation, except for T which is the ROHFSCF reference configuration function. Formulae for the Matrix Elements of PFDCIs
Again in the interest of compactness we shall express the matrix elements of the various types of spinadapted configurations, coupled through p, as the sums of the orbital densities they contribute to p,,. Let us start with the diagonal elements and recognize we are working with real spatial molecular orbitals, or Ii)(j = Ij)(il.
(ToIZITo) = Im)(ml (3.42) (T1I IT) = IJ)(JI (3.43) (T I IT,) = la)(al (3.44) (T3IFIT3) = Im)(ml (3.45) (T, ZIT,) = '(i)(il + Ia)(a  m)(ml) (3.46) now the offdiagonal elements of the type (TlpIT) where T, j T'
(TI IVZIT') = Ij)(J'i (3.47) (TIZIT') = Ia)(a'I (3.48) (T3l1ZT') = 0 (3.49)
84
(T = i)( if a = a'
(3.50)
= jla)(a' if i = i'
now the offdiagonal elements of the type (Tm, IT,)
(Tol IT) = Ij)(ml (3.51) (To.I'IT) = Im)(at (3.52) (ToI.ZIT) = 0 (3.53) (To0'z T.) = 'li)(al (3.54) (T P IT,) = 0 (3.55) (TlpZ IT,) = 'Im)(aI if j = i3 (3.56) (Tlp IT,) = 'Im)(a,l if j, = i, (3.57) (TI2 1iz ITI) =  1i2)(mI (3.58) (T21'ZIT) = ,li,)(mI (3.59)
(T3lpZIT,) =  Ii3)(i,1 if i3 $ i4 and a3 = a,
3= la)(a,I if i3 = i, and a3 $ a, (3.60) = ( (a)(al  Ii)(iI) if is = i4 and a, = a, This leads to
(P(rN)) = CCJ P"(x,6(r  rN)IX,) (3.61)
85
where I and J run over all configuration functions. The C, and C, are the CI expansion coefficients, equation (3.4), the indices y and v run over all atomic basis functions, {X}, and the P"' are obtained from equations (3.42) through (3.60).
Calculated Results for the ROHFCIS Method
As in Chapter 2 we again present results for several smaller test systems similar to those used in determining the optimum valence Slater sorbital exponent, in Table (D8). As was the case in Chapter 2, the INDO/1 Hamiltonian yielded consistently worse results than did the INDO/S Hamiltonian when compared to experiment.6 For this reason we limit our presentation of results to those of the INDO/S Hamiltonian. The Bacteriochlorophyll structures and the Bacteriopheophytin structures are those used in Chapter 2, and again all calculated results are presented in units of Gauss (G).
6. See Tables (D3) through (D5).
86
Table 32: Calculated results for Bchla and Bphea.
Magnetic Calculated System Center7 Positions a8,o a.,9 Bchla + N 21 0.09 0.78 N 22 0.47 1.12 N 23 0.52 0.83 N 24 0.08 1.03 H 5 0.35 0.84 H 10 1.46 0.46 H 132 1.36 0.58 H 20 0.08 0.46
Bphea  N 21 0.62 0.43 N 22 1.21 2.56 N 23 0.09 0.21 N 24 1.58 2.19 CH3 2' 2.47 2.53 H 5 3.28 3.02 H 8 0.57 0.65 H 10 0.83 2.85* CH3 12' 0.51 2.96* H 20 3.67 2.47 NH 21' 0.15 0.35 NH 23' 0.04 0.17
7. The isotopes are 14N and 1H.
8. See Figure (15)
9. Reference 77.
87
Table 33: The calculated results for Bchlb.
Magnetic Calculated System Center Positiono a,0o aexp 1 Bchlb + N 21 0.13 0.82 N 22 0.12 1.17 N 23 0.69 0.82 N 24 0.13 1.06 H 5 0.24 0.89 H 10 0.64 0.43 H 132 1.43 0.43 H 20 0.23 0.43
Comparing the a.,, and calculated ai,o columns of Table (32) and (33) we immediately see that the ROHFCIS results are good. The computed sign of the hfc are all correct. Unlike the UHF and PUHF methods the computed magnitude of the hfc are all reasonable. If we bear in mind that the experimental hfc were obtained in solution, and the structures are from xray crystallographic methods, then these results do indeed seem reasonable. We should also note that the * values are considered to be suspect by those who took the measurement.
While it required obtaining the eigenvector associated with the lowest eigenvalue of a matrix of order 26000, these results seem worth the effort.
10. See Figure (15)
11. Reference 77
CHAPTER 4
CONCLUSIONS
Several points need to be made before we can draw our final conclusions. Whether the tools we use to investigate nature are totally empirical, semiemperical or as close to first principals as the situation will allow, there is only one real measure of success; and that is, will our model describe observed reality? This leads to the following question, what is observed reality? Or stated in a less philosophical manner, how good are our measurements of nature? The qualification of this question will shade light on the "goodness" of our model.
Our model treats a single molecular system as the sole inhabitant of its universe. It interacts with nothing but itself and its conformation is fixed. Compare this to the world the experimentalists' molecular systems occupy; most experimental observations are made in the solid of liquid phases. (All molecular systems used in this work as measures to obtain our parameters, or as yard sticks to measure the parameters' "goodness", were themselves measured in the liquid phase.) In the few cases where both gas phase and condensed phase ESR spectra are available for the same radical species, the hfc typically differ by a few percent, but can differ as much as 10+ % for the extreme cases.
Computed properties such as spectroscopy, dipole moments, transitions moments, 88
89
etc. as well as ESR hfcs are more sensitive to structural changes in the molecular system than is the computed energy of the system. The hfc are particularly sensitive to such changes [117, 118]. Bartlett et al [119] in a work to be published, show this quite convincingly for triatomics, at several levels of theory. The structures we used in the calibration of our model were not all determined experimentally, many were obtained via quantum chemical techniques. The conjugated systems were optimized via an INDO method, where practical experience has shown the resultant structure to be as good as, or better than, those obtained via abinitio SCF methods. The alkane structures were obtained using abinitio methods, as the INDO method we use has known deficiencies when applied to such systems. In either case it is physical intuition and low level quantum chemical methods by which the structures used in this work were obtained. Others who work with smaller systems use methods which are capable of producing more precise structures, usually gas phase structures; but all in this business must ask, "is the structure we are using the same as that the experimentalist is measuring"? The bacteriochlorophyll and Bacteriopheophytin structures used in this work were obtained from the solid state, and the hfc , a,,, are from solution experiments. The "floppy" substituents of the Bchla, Bchlb and Bphea are prone to large displacements upon going from the liquid to the solid state, and visa versa. As noted above the hfc are very sensitive to changes in structure, so we chose magnetic nuclei that were substituents of the macrocycle or the conjugated
90
rings to use as measures in our calculations. Here we assumed that these rigid regions would be the least perturbed upon phase transitions.
Molecular vibrations effect the measured hfc and corrections for this are typically a few percent of the measured hfc value [118]. In extreme cases where large atomic displacements are possible, and especially when such displacements move the measured magnetic center out of a nodal plane into direct interaction with the orbital describing the unpaired electron, such as the case of the methyl radical, the vibrational averaging correction can be of the order of the magnitude of the observed hfc splittings. The differing slopes of Figure (10) may be partially due to such vibrations. Vibrational modes that would bring the hydrogen lying in the nodal plane, out of the plane and into direct interaction with the 7r molecular orbital describing the unpaired spin, would increase the unpaired spin density on the hydrogen. Though the effect would tend to reverse the sign of the spin density at the hydrogen; as the induced spin polarization at the hydrogen, due to the unpaired electron in the p, orbital of the adjacent carbon, would have a negative sign. Direct interaction between the hydrogen and the r electronic system of the molecule would place positive spin density at the hydrogen.
Multicenter exchange, which is missing from the INDO Hamiltonian, would of course, also effect the slopes of Figure (10), and this is something we will investigate in the future. A comparison of the CIS a and 7r 6,. of Tables (D6) and (D7) indicates
91
that, at least, for the systems included in the exponent optimization the slopes should be very close.
Figures (1214) are plots of a,,,, vs a,,,, for the UHF, PUHF and the ROHFCIS methods. The slopes of these plots are determined via a linear least squares fit and should equal 1.0; the intercepts should equal 0.0. The UHF and ROHFCIS slopes are close to 1.0 (i.e. 1.01 and 1.04 respectively), while the intercepts are respectively, 0.99 and 0.02. While the slopes for both methods are good, the intercept of the UHF method in not as good as that of the ROHFCIS method. The PUHF slope and intercept are both bad, compared to the other methods, and have values of 0.89 and 1.39 respectively. Since all data points plotted in these figures are for systems that were used to optimize (, these values are as good as the methods are capable of. From above, and as we shall see later the PUHF method is the least successful of the three, followed by UHF, and the ROHFCIS method will prove to be the best.
From Figure (10) we see that the two and threecenter terms are necessary. If we do not account for them we obtain two o, for hydrogen. One for the r systems and one for the 7r systems, the ,, w 1.48 while & g 1.20. Accounting for the two and threecenter terms leads to , = 1.248, and from Figure (14) we see that the inclusion of the twoand threecenter terms and using , = 1.25 does lead to good results for hydrogen. The aforementioned exponent being that of the ROHFCIS method, since both the UHF and

Full Text 
PAGE 1
A QUANTUM CHEMICAL APPROACH TO THE DETERMINATION OF THE SPIN DISTRIBUTION WITHIN LARGE MOLECULAR RADICALS By MARSHALL GEORGE CORY, JR. A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY . < UNIVERSITY OF FLORIDA 1994 ' 4 : ? '
PAGE 2
To my beloved mother Catherine, my father Marshall Sr., his wife Jeannie and my dear wife, Genny
PAGE 3
ACKNOWLEDGMENTS It is impossible to acknowledge all who have contributed to my University of Rorida experience. With intentions to slight none, I choose to mention a few who "stick out" for one reason or another. Â• Â• Dr. Michael Zemer, for his constant enthusiasm. Â• Dr. Charles Taylor; Charlie, the "a la personne qui a garde la porte du club au moment opportun" would not have been necessary if you and one other had not ran and hid like "wimps". Â• Dr. Ricardo Longo, my best friend at QTP and the best read graduate student I know of, or likely ever will know of. Â• Genny Cory, for putting up with this nutty endeavor over the years. Finally to all the friends I can remember at this moment, Rajiv and Priya, Ricky and Ivani, Quim, Steve and Trish, Christin, YaWen, Billy Bob, Chris, Mark, Dave, Zhengy, and of course Monique. To the small multitude of others, whose friendships I have enjoyed and whose names escape me at this moment, forgive me for not remembering in time. ill
PAGE 4
TABLE OF CONTENTS ACKNOWLEDGMENTS iu LIST OF TABLES vi LIST OF FIGURES vii ABSTRACT viii CHAPTERS 1 INTRODUCTION 1 The ESR Experiment 1 The Isotropic Hyperfine Interaction 5 Overview of SelfConsistant Field Theory 10 The BomOppenheimer Approximation 12 The Independent Particle Approximation 13 The INDO Approximation 25 2 THE UHF APPROACH 32 Introduction 32 The UHF Equations 33 Spin Contamination 37 Other Interesting Orbitals 39 Spin Contamination and the UHF PE Surface 44 Spin Projected UHF 46 The AUHF and EUHF Methods 55 Calculated Results for the UHF and PUHF Methods 57 3 THE ROHFCI APPROACH 61 Introduction 61 Restricted OpenShell HartreeFock 61 Configuration Interaction 63 Correlation Energy 65 Single Excitation Configuration Interaction 67 iv
PAGE 5
SpinAdapted Configuration Functions 74 Rumer Bonded Functions 74 Evaluation of for a Rumer CI Wavefunction 78 Fast Doublet CIS 81 FDCIS SpinAdapted Configurations 81 Formulae for the Matrix Elements of Ppdcis 83 Calculated Results for tiie ROHFCIS Metiiod 85 4 CONCLUSIONS 88 APPENDICES A UNITARY ORBITAL ROTATIONS 95 B THE MIRROR THEOREM 97 C THE PAIRING THEOREM 100 D THE EVALUATION OF V'Â„5(0)^ 103 The Optimum Slater Valence Exponent 108 The Charge Dependence of ^ 114 The Results of Optimizing the Valence Exponent 1 17 REFERENCES 124 BIOGRAPHICAL SKETCH 133 V
PAGE 6
LIST OF TABLES Table 21: The affects of annihilation on {S^) and calculated a.^o 55 Table 22: Calculated results for Bchla 58 Table 23: Calculated results for Bchli and Bphea 59 Table 31: How the correlation of the INDOROHF reference affects the calculated value of a,so 72 Table 32: Calculated results for Bchla and Bphea 86 Table 33: The calculated results for Bchl6 87 Table D1: The of the standard Nitrogen optimization data set, where = 1.95. 108 Table D2: The of the standard Hydrogen optimization data set, where = 1 .20. 109 Table D3: Minimization results for tt hydrogen 113 Table D4: Minimization results for crhydrogen 113 Table D5: Minimization results for 7rnitrogen 113 Table D6: The optimized hydrogen exponents 1 14 Table D7: The optimized nitrogen exponents 1 14 Table D8: Computed UHF, PUHF and ROHFCIS results for the optimum Slater valence exponent 115 Table D9: The effects of including the twoand threecenter contributions to a,Â„,,. 116 Table D1 0: The Optimization Data Set 117 vi
PAGE 7
UST OF FIGURES Figure 1: The electron magnetic resonance condition 3 Figure 2: Rumer bond diagrams for the case of three electrons in three orbitals and S'2 77 Figure 3: A closed loop and an odd chain 78 Figure 4: Two types of even chains which arise from (12) interacting with (1 (2 in the first case and (1 2) (3 4) interacting with (1 (2 3) (4 in the second. . 78 Figure 5: A multicycle closed loop, arising from the interaction between (1 2) (3 4) and (1 (2 3) 4) 79 Figure 6: The four possible orbital configurations for the FDCIS formalism 82 Figure 7: The Nitrogen Is probability density 119 Figure 8: The Hydrogenlike 2s probability density 119 Figure 9: The Slater type 2s' probability density 120 Figure 10: A plot of the spectroscopic CIS P^^^^ vs the experimental Aiso 120 Figure 11: The tt region of Figure (10) expanded 121 Figure 12: A plot of UHF a,,Â„ vs a^^, using = 0.961 121 Figure 13: A plot of PUHF a,Â„Â„ vs a^^, using = 1.389 122 Figure 14: A plot of aÂ„Â„ vs aÂ„p, using 65=1248 122 Figure 15: The number system for the Bacteriochlorophylls 123 vii
PAGE 8
Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy A QUANTUM CHEMICAL APPROACH TO THE DETERMINATION OF THE SPIN DISTRIBUTION WITHIN LARGE MOLECULAR RADICALS By MARSHALL GEORGE CORY, JR. April, 1994 Chairman: Michael C. Zemer Major Department: Chemistry Within the last decade the anticipation of some experimentalists for the ability to ascertain the structure of biologically important chemical systems has been realized. The elucidation of the structure of the photosynthetic reaction center of R. Viridis, including some 13,000 atomic centers, is a prime example, and this structure has led to a deeper understanding of the mechanisms of photosynthesis. Along with this increase in structural knowledge a concurrent increase in the knowledge, of the observable properties of such structures has also occurred. Of particular interest to this work are the magnetic properties of these systems, as revealed by Electron Spin Resonance (ESR) spectroscopy. The magnetic spectra, ESR, of such systems are quite complex and require some effort to understand, and a need for theoretical aids has been expressed. In many cases the ESR vill
PAGE 9
signal is the only structural signature available. This work has explored several quantum chemical methods for determining the isotropic hyperfine splitting (hfs) in such systems as those described above. Several semiempirical SelfConsistent Field (SCF) and postSCF approaches are explored. Two distinct parameterizations of the Intermediate Neglect of Differential Overlap (INDO) approximation are examined for their ability to produce good reference functions. Both Unrestricted HartreeFock (UHF) and Restricted OpenShell HartreeFock (ROHF) SCF wavefunctions are investigated, as well as the Configuration Interaction Single replacement (CIS) wavefunction. The results of applying the most promising method to several of the bacteria chlorophyll ions are reported and compared to those observed experimentally. ix
PAGE 10
CHAPTER 1 INTRODUCTION The ESR Experiment In general, magnetic resonance spectroscopy observes transitions between energy levels which depend on the strength of a magnetic field [13]. Experimental techniques which measure the interaction between an external magnetic field and the intrinsic electron spin magnetic moment [4, 5] are usually labeled electron spin resonance (ESR) spectroscopy experiments. Techniques which deal with interactions between orbital magnetic moments and an external magnetic field are labeled electron paramagnetic resonance (EPR) specuoscopy experiments. These terms are sometime used interchangeably; in this work we shall deal with the ESR problem as defined above. G While any atomic or molecular system containing one or more unpaired electrons will in general possess a magnetic moment due to the intrinsic spin of electrons, we shall be concerned in this work with molecular systems containing a single unpaired electron. (And as we shall see later, really only with those magnetic molecular systems that are "large" in the sense of abinitio quantum chemical theory.) This work is an attempt to develop a systematic and reliable method for the description of the spin distribution within large molecular systems. For radicals whose total electronic spin angular momentum S 1
PAGE 11
2 equals or in atomic units' ^, interactions such as electron spinspin coupling do not affect the experimentally measured spin distribution, and thus do not need to be addressed here and will contribute no fog to the overview of our final results. The vast majority of molecular systems have S values of zero; by far the next most prevalent value for S is ^. Seen in this light then, our stressing spin onehalf systems is not as constraining as it might seem at first, and if our attempt at developing reliable computational methods is successful then such multielectron interactions can be addressed at our leisure. Let us start with the simplest molecular ESR experiment. Consider a molecular system with a nonzero net electronic spin. Associated with this spin will be a magnetic moment. If a static external magnetic field is applied, the spin magnetic moment of the electron will interact with it. If the total electronic spin angular momentum S of the molecular system is ^, and the molecular system is spatially nondegenerate, then there are two magnetic energy levels available to the electron. These levels are degenerate in the absence of a second magnetic field. The degeneracy is lifted by the application of the static external magnetic field and the splitting is proportional to the external field strength (Hq). The basic ESR experiment then consists of observing transitions between these two levels, or states. 1. In atomic units h = e = rrie = 1, where h is Planck's constant divided by 27r, e is the magnitude of the electron charge and is the electron rest mass. Unless otherwise stated atomic units will be used through out the rest of this work.
PAGE 12
3 Transitions between the energy levels can be induced by the application of a second, appropriately oriented, external oscillating magnetic field. In our simple experiment the oscillating field is applied at a right angle to the static external field Hq. Transitions will occur when the resonance condition AE = hvQ = ge^eHo is satisfied. Figure (1) below. Figure 1: The electron magnetic resonance condition. Here h is Planck's constant (6.63 x 10"^^ erg sec), /3e is the Bohr magneton (9.27 x 10'^^ erg/G), ge is a dimensionless constant of proportionality equal to 2.(X)2322 for a free electron and i/q is the frequency of the perpendicularly applied oscillating magnetic field. The oscillating magnetic field has the same probability of inducing an upward transition as a downward transition. In a bulk sample at equilibrium in the static field the lower level will be more highly populated than the upper one, thus there will be more upward than downward transitions, and a net energy transfer to the bulk sample. It is
PAGE 13
4 this absoqjtion of energy that is detected, or observed, experimentally. If this was all there is to an ESR experiment then little useful information would be obtained by its being carried out. We would be able to obtain the constant of proportionality g, which would equal the ratio j^If for the moment we disregard the possible dynamic information intrinsic in the absorption line width and shape, then this would be the only piece of information we could extract. Fortunately, this simple picture is not complete. The simple system we used as our example above is not to be the case for the vast majority of magnetic molecular systems. Almost certainly there will be other intrinsic magnetic fields within the molecular system. These can be due to other unpaired electrons or orbitals which possess a net angular momentum and probably most importantly, to magnetic nuclei within the system. The interaction between the unpaired electron and the various magnetic nuclei^ gives a wealth of information on the distribution of the unpaired spin within the system; this in turn yields detailed information on such varied subjects as the molecular geometry, the electronic state of the system, the net spin of the magnetic nuclei can be determined, and from this the isotope of the magnetic center being observed. The experimentalist can also detect from the electronnuclear interactions the presence within the system of nuclear spinspin coupling as well as a wealth of other information we need not detail here. 2. A rather complete list of the magnetic nuclei and their various spin values can be found in Appendix (1) of Reference 6.
PAGE 14
5 From this brief overview it is obvious that magnetic resonance spectroscopy, for this work ESR, is a powerful tool for sciences' use in the investigation of the invisible world of atoms and molecules. It is just this power to yield, sometimes overwhelmingly so, detailed and complex information on the system being observed that spawned the need for, and our desire to develop, theoretical methods to aid in explaining, and hopefully predicting, the observed spectrum. The approximate molecular Hamiltonian^ is a thirtyfive term construct, some twenty of these terms will influence the experientially observed ESR spectrum. This dissertation shall primarily be concerned with only one of these, the isotropic hyperfine interaction, or the Fermi contact [8], a,so. The origins of the Fermi contact term is explained in terms of a magnetic dipoledipole interaction between the magnetic moments of an electron and a nucleus. The vector potential [9] due to the nuclear magnetic dipole moment is The Isotropic Hyperfine Interaction (1.1) where we use the standard mks units. The magnetic field due to fl^ is 5(r) = V X A(r) = ^(1.2) The interaction energy between jiff and /?e is 3r (f /r^) f^N f^e _ 3(r Â• /7e) (r Â• /T^) .3 ^5 (1.3) 3. See Appendix (F) of Reference 7.
PAGE 15
6 where f represents a vector joining the electron and nuclear centers, /Te and fl^ are, respectively, the electron and nuclear magnetic moments. The Hamiltonian for this interaction is written as Hdip = 9e^e9N^N SI^ 3(5r)(/^r) (1.4) where the collection of constants preceding the operators are, respectively, the free electron "gfactor" or Lande gfactor ge, the Bohr magneton ^e. the nuclear gfactor and the nuclear magneton 0^The operators 5" and / are the total angular momentum operators for the electronic system and the nucleus, respectively. When the electron is found in an orbital with an angular momentum of zero, i.e. / = 0 or an s state, the wavefunction is not zero at the nucleus. (In the Dirac theory [4] the atomic radial function diverges, but the divergence is so small that equation (D.7) is approximately correct for the nuclear dimension.) In any case ^, or ^, of equation (1.3) diverges at the origin. Â• / Suppose we take a small sphere with the nucleus as the center. Then the contribution to the isotropic interaction from the s electron outside the this sphere is zero, since the integral of pfor this volume stays finite. If the sphere is small enough, the electron density for the s electron inside the sphere can be regarded as a constant, and be replaced by the wavefunction at the origin, i.e. as V'(0)p. Since the s electron has a magnetic
PAGE 16
moment /?e, this small sphere is homogeneously magnetized with a magnetization of (1.5) The magnetic field, Hs, at the center of such a homogeneously magnetized sphere is M (1.6) or B = 4 IT (1.7) at the nucleus. The interaction energy between this field and the nuclear magnetic moment IS (1.8) (This is the FermiSegr6 term [10] and was first given by Fermi [8].) In operator form equation (1.8) is written as (1.9) where a,so equals StT o Zf \ Oiso = ^gePegsPsP (Tn) (1.10) Here />^(r;^) is the unpaired spin density at the center of nucleus (N). As we discussed earlier, an unpaired electron in the presence of a magnetic field has two states available
PAGE 17
8 to it, either aligned with or against the field. These states usually are labeled spin up and spin down or as alpha (a) spin and beta spin. The unpaired spin density at a point in space, p^{r Â— Vo), is then defined as the difiference between the a and ^ spin densities [11] at that point, or as it applies to equation (1.10) above = pÂ«(r,) /K) (1.11) The product ^^^gN^s is a constant. The constant of proportionality is not altogether constant. Nor can it in general be considered a scalar quantity. It is correctly expressed as a second rank Cartesian tensor [12]. Five terms contribute to the tensorial components [7]. Generally the largest contribution to the anisotropy of is through the oneelectron spinorbit interaction [13] Hso = <;LS (1.12) where L is the orbital angular momentum operator and {<;) is the spinorbit coupling constant. The oneelectron spinorbit interaction is proportional to the atomic nuclear charge raised to the fourth power, (Z^), [13] and becomes important for the heavier atomic systems. The molecular systems to which the methods developed within this dissertation are to be applied, are themselves constructed from the first, second and occasionally the third period elements. The oneelectron spinorbit interaction in these atomic systems is almost always quite small. The change in the average value of ge
PAGE 18
for these systems is typically at the third decimal place, an error smaller than the error intrinsic to the computational methods we will use to determine the spin density. Thus, for the methods developed within this dissertation the electronic gfactor will be treated as a scalar constant. This leaves . , _ (1.13) as the quantity to be determined computationally. Before describing the general details of how we go about computing equation (1.13) we need to present an overview of the quantum chemical approach we have chosen to use. Our stated objective is to derive a method that can, to predictable limits, determine the unpaired spin distribution throughout biologically important, and usually quite large, molecular systems, and to test it through the calculation of an observable property, a,so. This constraint as well as the desire to start out on known theoretical ground leads one to, as least in my opinion, semiempirical molecular orbital (MO) methods. Semiempirical MO techniques such as the various oneelectron Hiickel methods [14, 15] or the twoelectron ZDO [16] approaches have been applied to such systems in the past [17, 18] with mixed results. Their simple form and low computational cost make Hiickel MO methods logical first choices. Collectively they do an adequate job of describing the gross overall spin distribution of systems such as the porphyrins or the chlorophylls, systems that shall concern us later. On the other hand the calculation of quantitatively acceptable results
PAGE 19
10 by such methods has not been demonstrated. The assignment of the isotropic hyperfine splittings (hfs) through the use of the McConnell relation [1] and Hiickel spin densities have proven to be qualitatively useful. We expect to do better than this and obtain results that can be quantitatively useful. The aforementioned semiempirical Hiickel methods do not treat the electronic twobody interactions of a molecular system explicidy. The ZDO semiempirical methods we choose as our theoretical starting point do. It would seem from the start something of importance is missing from the oneelectron Hiickel methods. Also, primarily for the reason of the ignored electronic twobody interactions the Hiickel methods do not lend themselves to post SCF procedures, techniques we will, in later chapters, find to be of importance in obtaining reasonable values for the spin density, (1.13). Overview of SelfConsistant Field Theory Today the abinitio^ molecular orbital theory of choice is the HartreeFock SelfConsistent Field method [1925] (HFSCF). The ZDO theories we have chosen to use as our starting point are themselves of the HartreeFock selfconsistent field type. As such, it is necessary at this point to outline HartreeFock theory before introducing the ZDO methods. It is not our intent to develop the HartreeFock equations in detail as this lies outside the scope of this work. We do however need to give sufficient detail here so 4. Abinitio theories tend to make only those approximations which are necessary to keep the problem soluble.
PAGE 20
11 the ground work will have been laid for the methods discussed in later chapters. These methods will address the openshell problem direcdy. Here we will give an overview of closedshell HartreeFock theory, and leave the explanation of the various openshell approaches to each specific chapter. What we would like to be able to solve is the nonrelativistic, timeindependent, manyelectron Schrodinger equation, = (1.14) for the wave function ^, where H is the Hamiltonian for the system of electrons and nuclei and E is the energy of the system. Explicitly H is written as Â«=1 A=\ ^ i=\ A=\ i=\ i<3 ' A=1A
PAGE 21
12 two interacting body system, or a system that can be reduced to two interacting bodies. The problem lies in that the motion of the electrons and nuclei cannot be uncoupled for a three or more interacting body system. If we make the following two approximations we can imcouple the motions of the various particles and at least within the validity of the approximations solve for the wave function These approximations are known as the BomOppenheimer approximation and the Independent Particle approximation. The BomOppenheimer Approximation If we invoke the physical picture of the electrons moving among the nuclei so much more rapidly than the nuclei themselves move that we can consider the nuclei to be fixed with respect to the electrons, then we have also invoked the BomOppenheimer approximation [26]. This uncouples the motions of the nuclei from each other as well as from the motion of the electrons. The effect of this uncoupling on the form of equation (1.14) is as follows; the wave function *(ri,i?^) now parametrically, rather than explicitly, depends on the nuclear coordinates and we can write it as ^{rÂ„R^) = ^^^{r,)n{R^) (1.16) where ^(i?^) is the wave function describing the now separate motions of the nuclei and (1.17)
PAGE 22
13 is deemed the electronic wave function describing the electrons in the field of the "fixed" nuclei. Treating the nuclear positions as fixed parameters leads to M ^ and M M ^ ^ P ^ " Enucrep = static value (119) A=1A
PAGE 23
14 electronic coordinates) and no closed form solution can be obtained. However, equation (1.21) can be solved for the oneelectron case which leads us to a crude approximation. We ignore the electronelectron interactions and solve equation (1.21) as N oneelectron problems, where N is the number of electrons in the molecular system. The total Hamiltonian for the A^electron system within this independent particle approximation would become ; , ^ = ^1 + ^2 + ...'^^ vv(1.22) with ' . (1.23) a sum of oneelectron Hamiltonians, equation (1.22). This form of the Hamiltonian leads to a wave function of the form * = Xi(rOx2(rOXN(r;.) (1.24) a product of oneelectron fimctions, orbitals, and the total energy of the system will be the sum of the orbital energies. E = + Â• Â• (1.25) where the {x} in equation (1.24) are hydrogenlike wave functions or an approximation to them.
PAGE 24
15 Assuming that the oneelectron Hamiltonian, equation (1.22), and the wave function, equation (1.24), are correct we can now reintroduce the electronelectron interactions as corrections. This is done in an average way, {im = J Jdr, dr, x:(i) XU) jXi{i) xAJ) (126) which represents the repulsion between two electronic charge distributions, xi(OP Xj(i)p. Adding these averaged electronelectron interactions back into the oneelectron Hamiltonian yields the twoelectron Hamiltonian. Hartree proposed the SelfConsistent Field (SCF) method in 1928 [19]. He assumed the repulsion experienced by a particular electron is a spherically symmetric average of the potential due to the 1 other electrons. This leads to the replacement of the exact twoelectron potential of equation (1.27) with an effective oneelectron potential ^ = E^/^' = EE^^w (1.28) ; f where Jj{i), the Coulomb operator, is defined as Jj{i)= f dr,^^^^^^^ . . (1.29) When Jj operates on another charge distribution it yields the classical interaction between two charge distributions. It is now possible to express the total /^electron Hamiltonian
PAGE 25
16 as a sum of effective oneelectron Hamiltonians of the form = hi + v;ff (1.30) leading to the set of pseudoeigenvalue equations which determine the oneelectron wave functions {i}. Hartree then assumed a set of wave functions, orbitals, and proceeded to calculate the Hartree operator for the ith electron by evaluating the integrals of the effective potential. The resulting N equations are solved for new orbitals which are then used to compute the Hartree operator for the next iteration. This process is considered finished when the i + 1 iteration is within a given tolerance of the ith iteration at which time the potential is termed selfconsistent. Hartree' s SCF method failed [27, 28], this was due primarily to the form of the wave function. It is a fundamental principle of quantum mechanics that the measurement of any physical property which depends on the coordinates of a system of identical particles should not depend in any way on the arbitrary labeling of the particles. For two electrons described by the wave function ^( x.(l) X>(2) ) there exists a probability density (1.31) (1.32) which cannot depend upon the permutation of the electron labels. I*(X.(1)X.(2))' = *(X.(2)X,(1))P (1.33)
PAGE 26
17 Which leads to two possibilities x.(l) X;(2) ) = X.(2) X.(l) ) (1.34) or X.(l) X.(2) ) = X.(2) X.(l) ) (135) the symmetric, equation (1.34), and antisymmetric, equation (1.35), cases respectively. Experimental evidence [29] indicates it is the antisymmetric^ case to which the wave function for two electrons must conform. Hartree's simple product form for the wave function is symmetric with respect to coordinate interchange and this is the principal reason for the failure of the Hartree method. Working independently of one another both Fock [21] and Slater [22] applied the variation principle to Hartree's method. By varying the radial functions of the simple Hartree product to minimize the total energy they rederived Hartree's original intuitive method on a more rigorous footing, and assured the total energy so obtained is an upper bound to the exact ground state energy. When the additional constraints of maintaining orbital orthonormality during the variation, along with that of an antisymmetric wave function are applied to Hartree's method, the HartreeFock equations are obtained. The constrained minimization is 5. It is a postulate of quantum mechanics that the wave function describing systems of indistinguishable particles, which themselves possess half integer spin, must be antisymmetric with respect to interchange of particle labels.
PAGE 27
18 usually carried out using Lagrange's method of undetermined multipliers. While the antisymmetric wave fimction is constructed by the application of A on the simple Hartree product. The operator A is the antisymmetrizer. P The summation runs over all A^! possible permutations of the particles, p is the parity of the permutation and is equal to the number of twofold permutations needed to bring the orbital product back to normal order. It was Slater who first noted that equation (1.36) was the prescription for the expansion of a determinant [30]. This leads to wave functions constructed from antisymmetrized products of orbitals being commonly referred to as Slater determinants. The HartreeFock equations are given as N ni)M^) = Yl^bM^) (137) 6=1 where / is the Fock operator N 1 /(I) Ml) Mi) + E(W)^6(i)) 6=1 <^a(l) (1.38) The operator J of equation (1.38) is the same Coulomb operator as that of Hartree's method. The operator K is the exchange operator and gets its name from the exchange of electronic coordinates that takes place during its operation. The exchange interaction in the HartreeFock equations arises due to the antisymmetric nature of the electronic wave
PAGE 28
19 function, and so did not occur in the Hartree method. The Coulomb operator was given previously as equation (1.29). The exchange operator is defined as ^ . Ki{l) Ml) = y dr, m) <^a(2) hil) (1.39) Equation (1.37) can be brought into the form of a standard eigenvalue problem by a unitary transform of the matrix of Lagrangian multipliers. /(I) <^Â„(1) = Â£a<^a(l) ' ' (1.40) The total electronic energy as obtained from the HartreeFock method is given by N N N I i j>i where *o) is the Slater determinant formed from the N occupied HartreeFock orbitals. Originally the HartreeFock method was applied to atoms and the resulting differential equations were solved numerically. For molecular systems this is impractical for all but diatomics. The difficultly arises in the calculation of the HartreeFock orbitals or Molecular Orbitals. Roothaan [23] and Hall [24] both suggested that the HartreeFock orbitals be expanded as a linear combination of some known set of basis functions according to where is a fixed atomic orbital (AO) basis function and C^i is the expansion coefficient for the i/th atomic orbital in tiie ith molecular orbital. This, Linear Combination
PAGE 29
20 of Atomic Orbitals Molecular Orbital (LCAOMO), linear expansion reduces the numerical solution of the HartreeFock equations to an algebraic one. The HartreeFock problem becomes a system of linear equations which are solved for the expansion coefficients C^iThe LCAOMO expansion has the added advantage of providing molecular orbitals with a definite analytic form, a huge advantage over numerical orbitals when calculating properties. Substituting equation (1.42) into equation (1.40) yields V V left multiplying (1.43) by xjl and integrating over all electronic coordinates yields ^t^^^^i = SiipC^iei (1.44) V V where F^v = J dT.xlil) fXuil) (1.45) V = / dr.xlil) X.(l) (1.46) If there are atomic basis functions then there will be molecular orbitals and N equations of the form (1.43). In matrix notation equation (1.44) is written as FC = SCE (1.47) and is often referred to as tiie RoothaanHall equation. This is the equation we solve in an iterative manner until selfconsistency is achieved. Upon convergence the columns of
PAGE 30
21 C are the coefficients for the AO expansion of equation (1.42), one column or MO per original AO, and the eigenvalues E are the HartreeFock MO energies. At this point we need to express the closedshell spin Restricted HartreeFock (RHF) equations in their fine detail. This will allow for a better understanding of the ZDO based approximations to them. Also, up to now we have been cheating a little. That is, we have not addressed electron spin. In 1925 Uhlenbeck and Goudsmit [31, 32] proposed the electron has an intrinsic angular momentimi, or "spin", to explain the results of the Stem and Gerlach experiment [33, 34]. Therefore our description of the electron using three spatial coordinates is incomplete, a forth coordinate, the spin,^ is needed. It was postulated that the magnimde of this spin was ^ unit leading to eigenvalues for Sz equaling It is customary to designate this spin coordinate u and to introduce two orthogonal spin functions a{uj) and /3{u;) such that {a\a) = W) = l (1.48) {a\^) = (^a) = 0 (1.49) 5a) = 5(5 + l)a) = fa) (1.50) m^S{S + im = i\/3) (1.51) 6. Dirac placed the intrinsic electron spin on a solid physical footing with his relativistic treatment of the electron in 1928 [4].
PAGE 31
22 Sz\a) = l\a) (1.52) 5.1/9) = (1.53) If we multiply the set of molecular orbitals {}, or the set of atomic orbitals {x}, by these spin functions two sets of "spin" orbitals are obtained. The properties of the spin orbitals are such that tki = (1.54) ^i = i\/3) (1.55) (V'.V'.) = (M{m = 1 (1.57) (V.l^.) = = 0 (1.58) or two sets of spin orthogonal orbitals. How the inclusion of spin into the wave function affects its construction will be explored in detail in chapters two and three, equations (1.54) through (1.58) above are sufficient detail for now. The introduction of the spin functions a{uj) and /3{u}) into the electionic wave function has the effect of enforcing tiie Pauli Principle. Two elections can be described by, or occupy, one spatial orbital only if they are of different spin. In Restricted HartreeFock tiieory all occupied molecular orbitals are doubly occupied; there are no half filled, or openshell, orbitals in RHF tiieory. Hence die term
PAGE 32
23 closedshell. If we take our two sets of spin orbitals, and {ip}, there will be y occupied orbitals in each set, where N is the number of electrons. If we expand the spin molecular orbital basis as was done in equation (1.42), this time using the spin atomic orbital basis, again substituting the constructs into equation (1.40), and integrate over the spin coordinates, the restricted closedshall HartreeFock operator is obtained. N/2 /(I) = hil) + [2Ja{l) K^il)] (1.59) a where Ja{l)= I dr,:{2) Â— M2) (1.60) Kail) = I dr, l{2) ^ U2) (1.61) The operator % is a twofold permutation operator, operating on the electron labels. Its effect on the simple orbital product (f>i{l) (t>j{2) is p,3<^.(l)^j(2) = ,^i(2)<^,(l) (1.62) Pi2 exchanges the electronic labels and the permutation operation takes place prior to the integration. The factor of two in equation (1.59) stems from the fact that an electron in one molecular orbital will have two Coulombic repulsions and one exchange interaction with the pair of electrons occupying a second, different, molecular orbital. There is exchange
PAGE 33
24 only between electrons of the same spin. The energy of a RHF Slater determinate is (1.63) E = 2^{a\h\a) + Y^[2{ah\ah) {ah\ha)] = 2^/iÂ„Â„ + ^[2jÂ„i/i:Â„6] a 0,6 The /th orbital energy is given as ^ ei = hii + YfiJiaKi,] (1.64) We are now in a position to expand the closedshell Fock operator matrix elements into the atomic orbital basis. It is from this vantage point that we will be able to best understand the effect of the various ZDO approximations on HartreeFock theory. The matrix elements of the closedshell Fock operator are given by N/2 Wlu) = {,i\h\l.) + Y^[2{fi\Ja\u) {fi\Ka\l^)] (1.65) a where the Coulomb matrix elements for the ath molecular orbital are Walu) = / dr, <^:(2) Â— <^a(2)i/(l)) = (/iGi/a) (1.66) and for the exchange operator {fi\Ka\iy) = (/x(l) [ dr, ct>l{2) Â— <^a(l)k(2)) = (/.aaz/) (1.67) where in equation (1.67) %2 has already operated. The notation is as follows: n and V are atomic basis functions while a is a molecular orbital. Replacing Ja and Ka in
PAGE 34
25 equation (1.65) with equations (1.66) and (1.67) respectively gives ffiv = hfiv + ^[2(//az/a) {na\av)] (1.68) a an expression containing both atomic and molecular orbitals. Expanding ^q. in equation (1.68) above, in the manner of equation (1.42) yields Nl2 = V + E E CAaC:j2(/.<7i/A) {,.a\Xu)] (1.69) N/2 = V + 2 E E ^Aac:j(/.^kA) iHXi.)] a aX = V + E ^A(T[WkA) hfi(T\Xl/)] where the Pa
PAGE 35
26 In words, for X/x 7^ Xf the differential in equation (1.71) is defined to equal zero. Thus the matrix of atomic overlap integrals becomes the unit matrix. The effect of this approximation on the oneand twoelectron integrals, in terms of the number of such integrals and the physics they contribute to the theory, is pronounced. All three and four center Coulomb integrals are removed, and no exchange integrals remain, or ^liAi'B ~ ^/if (172) {^^A>^c\l^B(^D) = 6ftÂ„6xa{fiX\iJiX) (1.73) Throughout this introduction we have made reference to the "various" ZDO methods, we are now at a point where we can at least list them, and outline the one pertinent to this work. Today there are numerous ZDO methods in use. They are all in one way or another based on of the following approximations, the Complete Neglect of Differential Overlap (CNDO) approximation [35], the Intermediate Neglect Differential Overlap (INDO) approximation [36], and the Neglect of Diatomic Differential Overlap (NDDO) approximation [35]. The CNDO method is true to the ZDO approximation, while the INDO and NDDO methods relax somewhat the ZDO constraint on the twoelectron atomic integrals. All SCF calculations performed within this thesis were carried out at the INDO level of approximation and so it is the INDO method we will outline below.
PAGE 36
27 Within the INDO approximation the RoothaanHall equation, relation (1.47), is constructed and iteratively solved as was previously outlined. The difference between INDO and abinitio HartreeFock lies in the type of integral included, and their functional form, in the construction of the Fock matrix, with two caveats. The first being that the INDO method described here explicitly uses only the valence atomic orbitals, and the second being that in the LCAOMO expansion only one basis function is used for each atomic orbital. Thus we will be describing, and working with, a valence orbital minimum basis set INDO method. Below the INDO equations analogous to equation (1.69) are presented, //, v, a and A again label the atomic basis functions, {x}, while A,B,C, , label the atomic center the AO basis functions are associated with. The imposition of the INDO approximation on equation (1.69) leads to different forms for specific cases, these are listed below 1. Case fi = v; A = B: FfiAliA = t^/t^/lA + [Pec IaC Vj,c\ (1.74) Xa <^A 2. Case n ^ u; A Â— B: ^/ix/ix = J] ^"^^y^ [(/^aA^ Ii^x^^a) \{iiaK Wa^a)] (1.75) \a <^a
PAGE 37
28 3. Case \i ^ v\ A ^ B\ ^t^AVB Â— l^tlAfB ' 2 ^HAVbIaB (176) Several new terms were used in expressing equations (1.74) through (1.76), these are the terms used throughout the literature. We shall explain them below starting with the onecenter core integral (/^^^^ of equation (1.74), C^Mx/*x = (^xiv2 + yj/z^) (1.77) the onecenter electronnuclear attraction, V^ = Â— (1.78) the first order atomic density Pec is defined as, Pcc = Y.P^cXc (1.79) Ac the twocenter twoelectron is defined as, Iac = (/^A Ac 1/^.1 Ac) (1.80) and Vac tiie twocenter electronnuclear attraction as Vac = {iIaWcIiIa) (1.81)
PAGE 38
29 The I3fi^Â„g term is explained as replacing ^^Ai^B = if^A V,V,\i^g) (1.82) A book could be written, and many have been [16, 3739], on the subject of obtaining the various terms above. The details of this subject lie outside the scope of this thesis. It should be said the various integrals are replaced by functions of experimentally obtained parameters such as ionization potentials and electron affinities, with the exception of in equation (1.82). The /3 term is, in many implementations, a totally empirical parameter and is usually obtained by varying it, while calculating a specific property or reproducing results of a higher level of theory, until acceptable results are obtained. Such is the case in this work, we shall obtain the SCF orbitals via two INDO parameterizations, one having been parameterized to reproduce molecular structures and the second to reproduce the low energy ultraviolet and visible absorption spectra of molecular systems. Both parameterization follow the philosophy of Pople's original lNDO/1 [36] method, at least with respect to the functional form of the U and 7 integrals and in the sense of varying /3 to reproduce a benchmark value. ' We are now in a position to detail how one actually evaluates the spin density, (1.13). At this point we will limit the discussion to a single Slater determinant. The point was made earlier that the unpaired spin density at any point in space (r) was defined to be the alpha density at point (r) minus the beta density at point (r). In terms of the spin
PAGE 39
30 molecular orbitals, equations (1.54) and (1.55), the alpha and beta densities evaluated at (r) are expressed as N" P"ir) = IC(r)P (1.83) a and TV" /(r) = E \'^a{r)f (1.84) a Then ^^(r) = ^"(r) Â— p^{r) and if we expand equations (1.83) and (1.84) in the manner of equation (1.42) we obtain P^i^) = T.T.P>'^(^)x:{r) (1.85) /i 1/ and ^'w^EE^ifx.wxrw (1.86) where P*^ and P^v are again elements of the first order density matrices of the alpha and beta spaces respectively, or N" p^,. = Y.^uc^.ay (1.87) a . PS. = J2^UCSa)\, (1.88) a Again the CftÂ„ are the expansion coefficients obtained from the solution of the RoothaanHall equation, relation (1.47). The expression for p^{r) can now be written as P'i^) = E E PS.)x.{r) Xlir) (1.89)
PAGE 40
31 where all quantities in equation (1.89) above have been previously defined. We should note that for the closedshell RHF theory we previously outlined, the term ^P*^ P^^^ is everywhere identically zero, a consequence of each spatial orbital being doubly occupied. We can now proceed with the thrust of this work, that is, to develop a method capable of quantitatively describing the unpaired spin distribution throughout a molecular system as measured through the calculation of the isotropic hyperfine interaction, aiso, and to be able to apply this method to systems that are of interest to chemists today. In particular, we will eventually be interested in very large systems, and so our aim will be more toward applying these methods to semiempirical quantum chemistry. However, the theory we develop should, in general, be appropriate for any wave function.
PAGE 41
CHAPTER 2 THE UHF APPROACH Introduction In Chapter 1 we developed RHF theory in sufficient detail as to allow for the formal definition of p^{r). Unfortunately, RHF theory, by construction, cannot address molecular systems which possess unpaired electrons. The simplest method for addressing such systems at the SCF level of theory is the Unrestricted HartreeFock (UHF) method which is also known as the Different Orbitals for Different Spins (DODS) [4045] approach. Historically this method dates to the late 1920s where both Hylleraas [46] and Eckart [47] applied it to the He atom, and the method as it is formulated today dates to the early 1950s. Within UHF theory electrons are still assigned to orbitals in Slater determinants [36], but the spatial orbitals associated with aspin electrons are allowed to differ from those describing the /3spin electrons. This additional freedom has several consequences; due to the increased number of variational parameters the UHF openshell energy will be < the ROHF^ energy, as well as allowing for a more accurate treatment of electron correlation at the SCF level [4850]. Perhaps the most important ramification, at least for this work, is that a single Slater determinant constructed from UHF spin molecular orbitals 1. Both the Restricted OpenShell HartreeFock (ROHF) method and the concept of electron correlation will be treated in Chapter 3. 32
PAGE 42
33 r.:' will not in general be an eigenfunction of [51, 52], though it will be an eigenfunction of SzAs we shall see later in this chapter this so called "spin contamination" will prove to be a problem in the calculation of aiso in the larger radical, molecular or ionic, systems. We now need to develop UHF theory in sufficient detail to understand how Puhf{^) determined. The UHF Equations Â«Â• In Chapter 1 we found the general HartreeFock eigenvalue equation to be f{l)rPa{l) ^ SaM^) . (2.1) where the {ipi} are the restricted spin molecular orbitals. We shaU use equation (2.1) as our starting point in developing UHF theory. In a manner analogous to that used in defining equations (1.54) and (1.55), the restricted spinorbital definition, we define two sets of unrestricted spinorbitals. ^i = (f>i\a) (2.2) i^^ = M (2.3) where the restriction that (f>^ = <^f imposed in RHF or ROHF theory has been dropped, thus giving the method its name. Unrestricted HartreeFock. If we now replace the {ipi} in equation (2.1) with the forms defined above, left multiply by (a and (/?, respectively.
PAGE 43
34 and integrate the resulting equations over the limits of the spin variable u we obtain r(l) = el 4>^{\) (2.4) and r(i) <^f(i) = Â£f i{i) ^ (2.5) where r (1) = h{i) + [^a (1) + E ^'(1) (2.6) r (1) = Ml) + E (1) ^6(1)] + E (1) (2.7) 6 a the aspin and /?spin Fock operators respectively. The summation upper limits p and q are respectively the number of aspin and ^spin electrons, and by convention p > q. Since the kinetic energy of the electron and coulombic nuclear attraction are independent of spin, h{l) is identical to that defined in equation (1.23). The unrestricted coulomb and exchange operators are defined below, the notation is the same as that used in defining equations (1.60) and (1.61). J^(l) = / dr, ra*{2) Â— ra{2) (2.8) J?(l) = / ^r, (2) ^ ^,^(2) (2.9) K^il) = / dr, <^r(2) Â— <^a(2) (2.10) J '"12 ^,^(1)= / dr,r:{2)^t{2) (2.11) J '12
PAGE 44
35 The energy of a UHF Slater determinant is given by f 1 r p 1 1 p a b a,b (2.12) a,b a. b where equation (2.12) is analogous to that given for the RHF energy of a Slater determinant, equation (1.63). The J^j term is defined as the coulombic repulsion between the aspin and ^spin electrons whose spatial distributions are described by \l\^ and jp respectively, aU other terms in equation (2.12) are analogous to their RHF counterparts. We now expand our two sets of spin molecular orbitals as a linear combination of some known set of basis functions (i.e. The LCAOMO approximation.) consisting of m linearly independent functions {x}. drjr, <^r(l)C(2) Â— KmiC^) (2.13) ^12 m (2.14) m (2.15) If we now substitute equation (2. 14) into equation (2.4) we obtain (2.16)
PAGE 45
36 now multiplying equation (2.16) by xjl(l) and integrating over yields Efa /la (2.17) V V where all quantities in equation (2.17) were defined for equation (1.44), and (2.18) In matrix notation equation (2.17) is written as (2.19) Similar results are obtained when one starts by substituting equation (2.15) into equation (2.5) and carrying out the subsequent integration, or These two equations, (2.19) and (2.20), are the unrestricted generalizations of the RoothaanHall equation, (1.47). They were first given by Pople and Nesbet [40], and are often referred to as the PopleNesbet equations. Once the matrices CÂ° and have been determined for a molecular system, /9^(r) is determined from the following. If A is a rectangular matrix of order m x p constructed from the first p singly occupied orbitals, columns, of C", and B is a rectangular matrix of order m x 9 constructed from the first q singly occupied orbitals, columns, of C, then (2.20) (2.21)
PAGE 46
37 P" = BB^ (2.22) the adensity and ^density matrices respectively, which are used to construct the total density matrix = P"^ + P^ (2.23) which follows directly from equations (1.87 1.89) and uiso is obtained from equation (1.10). Spin Contamination While the UHF method has the important advantage of directly producing and P^^ at the SCF level, it also has several drawbacks as far as this work is concerned. It suffers from the same problem any single determinant method has in describing an openshell system which is not of the highest multiplicity, (e.g. A singlet system consisting of two openshell orbitals describing two electrons.) This is not an insurmountable problem as the molecular systems in which we are interested rarely possess sufficiently high symmetry for there to be nonaccidental orbital degeneracies; of course, accidental degeneracies or near degeneracies will be a problem when they occur. The crucial problem lies in how "^uhf, not in general being an eigenfimction of S\ affects PuHri^)As we shall see later the affect on Puhf{^) can be sufficiently pronounced so as to render the calculated a,so both quantitatively and qualitatively useless. For now we need to
PAGE 47
( 38 present an overview on the origins of spin contamination so the sections dealing with minimizing its effects upon, or removing it from, Pujjp{r) will be more lucid. A single determinant UHF wavefunction is of the form = ^<^^(l)a(l) cf>;{p)a{p) <^f (p + l)^{p + 1) <(n)/3(n) (2.24) where in the basis of UHF canonical spin molecular orbitals imj) = im',) = (2.25) and in general mr,)^o (2.26) The functions {(j)"} and {^} can be transformed among themselves with unitary transforms Vi=J2rsUsi (2.27) Vi^Y^lVsi /> (2.28) and still leave (2.24) unaltered.^ , Now by applying S' = SS+ + Sl + S, (2.29) 2. See Appendix A. i
PAGE 48
39 to (2.24), it is readily shown that p 9 {^vhAS'I^uhf) ^Sl + S,+qY,Yl (2.30) Â« j Equation (2.30) is general and holds for any single determinant wavefunction. (e.g. RHF, ROHF as well as UHF.) It is immediately clear that for RHF and ROHF wavefunctions the last two terms on the righthandside (rhs) of equation (2.30) cancel, (i.e. The overlap matrix in the basis of restricted HartreeFock molecular orbitals is the unit matrix.) and RHF /ROHF Â— 5^(5^41) (2.31) However for the UHF wavefunction {S')aHF > S,iS, + 1) (2.32) where the equality in equation (2.32) occurs when the UHF wavefunction converges to an exact eigenfunction of S^, the RHF/ROHF solution, else UHF EXACT (2.33) Other Interesting Orbitals The unrestricted molecular orbitals are, of course, not the only interesting type of orbital. We shall find two other types of orbitals to be of use in understanding spin contamination in a UHF wavefunction and in removing it. These are known as the "natural" orbitals [43], which are those orbitals that diagonalize the reduced firstorder
PAGE 49
40 density matrix; and the "corresponding" orbitals of Amos and Hall [43] which are also known as the "paired" orbitals of Lowdin [53]. These are the orbitals which diagonalize the overlap matrix between the aspatial orbitals and the ^spatial orbitals. For the wavef unction (2.24) the firstorder density matrix is 7(11') = j2 ^r(l') + E '^'^(l) ^'{l') r (l)^(l') (2.34) r a This is already in diagonal form and, since the canonical spinorbitals are already orthonormal, they are the natural spinorbitals. The natural spinorbitals are not unique since the unitary transforms (2.27) and (2.28) leave the form of (2.24), and (2.35), unchanged.^ The reduced firstorder density matrix is obtained by integrating (2.34) over the full range of the spin functions a{u) and ^{uj), this yields =^r(l)<^^(l') + E'^r(l)Â«^f(l') (2.35) r s which would be in diagonal form except that {(f)"} and {4>^} are not mutually orthogonal. The unitary transforms (2.27) and (2.28) can be used to define new orbitals having the property that their spatial overlap matrix is diagonal,'^ or Trs = J drr^frjl [ (2.36) 3. See Appendix A. 4. When the unitary matrices U and V, of equations (2.27) and (2.28), are determined as in Appendix C.
PAGE 50
41 Tts = TrSrs (2.37) these are the "corresponding" orbitals of Amos and Hall. In terms of the corresponding orbitals (2.35) becomes Kill') = Era) <(!') + 1:^(1)^1(1') (2.38) r s where the {7/Â°} are spatially orthogonal to each other and to the {tj^} except when r Â— s. If these sets of paired functions are replaced with the Schmidt orthogonalized [54] functions K = {r,; + r]'^){2 + 2Tr)'^ (2.39) f^r = in; 2Tr)^ (2.40) then all the functions are orthonormal and (2.38) becomes /'(ii') = E(i+3^r)A:(i)A,(i')+E(i^^)/'**(i)/^^(i')+ E ^r(i)<^t(i') (2.41) the functions A^, ^ls and of relation (2.41), are the natural orbitals. The corresponding orbitals are then the natural spinorbitals most easily related to the natural orbitals, and the overlap integrals of the "paired" orbitals determine the natural orbital occupation numbers. A second approach can be taken in determining the UHF natural orbitals, one that will give further insight. First we express the reduced firstorder density matrix in terms
PAGE 51
42 of the expansion basis {x} m p(ll') =Y.(Prs+Pr's)x:{l)Xs{l') (2.42) r,3 where P^^s and P^^, are matrix elements of relations (2.21) and (2.22) respectively. To see the relationship between definitions (2.41) and (2.42) we first define the matrix C = [ABUÂ„ (2.43) where [AB] is the concatenated matrix formed by matrix A of relation (2.21) and matrix B of relation (2.22), mdn = p + q. Then from equations (2.21) and (2.22) it follows that CC^ = (P'"+P^)Â„xm ' (2.44) and from Appendix B this has the same nonzero eigenvalues as C+C. Now C^C=(h (2.45) where I, and I, are the unit matrices of order p x p and q x q respectively, and S is the overlap matrix between the UHF aspatial orbitals and ^spatial orbitals. When the unitary matrices U and V are determined as in Appendix C, relation (2.45) can be brought into the form v^)(s^ 0(o v)=e. I) where T is the diagonal matrix of corresponding orbitals overlaps. There are then 2q eigenvalues of the form 1 Â± Tr (2.47)
PAGE 52
43 and {p q) eigenvalues that are exactly unity, as in (2.41). If the number of expansion basis functions {x} is less than the number of electrons in the molecular system, (i.e. m < n), then there will be m n overlap integrals, equation (2.36), that equal unity. This corresponds to a set of orbitals whose occupation numbers equals 2. There will be n m = D of these doubly occupied orbitals, and they have no pairing properties. If m > n, there will be a set of orbitals which have occupation numbers of zero. There will hemn = E such orbitals, and they also have no pairing properties. When m = n, then all overlap integrals between the paired orbitals will be less than unity and greater than zero.^ i.e. n > m, D n Â— m (2.48) E 0 n < m D 0 (2.49) E m Â— n n = m D = E = 0 (2.50) These relationships can be stated clearly as the eigenfunction equation (2.51) 5. See Appendix C.
PAGE 53
44 where 'p is defined as P = Y,dkOk{\)ei{l') (2.52) *;=1 then Ok=rk = ril dk = 2^1 + Tk, k = l,2,D (2.53) ek = \k, dk = l + Tk, k = D + l,q (2.54) 0k = tik, dk = lTk, k = D + l,q (2.55) ek = r]l = (f>l, dk = l, k = q + l,p (2.56) There are no relations for the unoccupied orbitals {E}. Spin Contamination and the UHF PE Surface For RHF and ROHF wavefunctions the natural orbitals are the canonical molecular orbitals and their occupation numbers are either 2, 1 or 0. It then follows that for an exact spin eigenfunction occ ^aJ = 49 + (p9) (2.57) i where the o,are the natural orbital occupation numbers [55]. For a single determinant no c7, other than the RHF and ROHF values can satisfy (2.57) while conserving the total number of electrons. For a UHF wavef unction the natural orbital occupation numbers
PAGE 54
45 are functions of tiie corresponding orbital overlaps, or, (7i = l + Ti, i = l,2,q j" ... (7.= 1r,', i' = l,2,q (2.58) (7^ = 1, u^2q + l,2q + 2,p where i and i' are paired. Thus for a single determinant UHF wavefunction to be an exact spin eigenfunction tiie T, must equal 1, i.e. tiie corresponding orbital overlap matrix must be the same as that of an RHF or an ROHF wavefunction. This implies tiiat the UHF wavefunction must be related to tiie RHF or tiie ROHF wavefunction tiirough suitable unitary transformations. To this point we have expressed the Fock operator as a function of the Fock molecular orbitals. We could have just as easily expressed it as a function of the total density P"^, let /(pT) = h + J(P^) lK(P'') (2.59) ^(P") = + P'^) / ^*(2) ^ u{2) (2.60) K{Fn = J2(P^, + P'.) /rfr,/(2)^K2) (2.61) J '12 HI/ Then in the basis of the corresponding orbitals we can write the UHF equations as (^r l/(P^) iK{F')\r,) = 0 , i = 1, 2, Â• Â• Â• p, 1/ = p + l,p + 2, Â• Â• Â• m (2.62)
PAGE 55
46 {vf\f{Pl + lK{F'M)=0, i = l,2,9, u = q + l,q + 2,m (2.63) where = P^. For an ROHF like solution the {r]"} and {rj^} must be the same, or, (2.63) can be written as {Vi\f{P^) + ^K{F')\vt) = 0,i = l,2,p, u = p+l,p + 2,m (2.64) Comparing equations (2.62) and (2.64), it is obvious that they both cannot be simultaneously correct without P^ = 0, identically. This corresponds to an RHP solution, but from relation (2.57) an openshell spinexact single determinant wavefunction must be related to an ROHF wavefunction. Thus no single determinant spinexact wavefunction is stationary on the UHF potential energy surface. This result will have some bearing on the next two sections. Spin Projected UHF As noted above the single determinant wavefunction (2.24) is an eigenfunction of Sz with {Sz) = ^{pq). However it suffers from spin contamination, described previously, and for the aforementioned reasons is not an eigenfunction of S'^. Since we use the PUHF method to "improve" our UHF wavefunction, we will describe below this method (spin projection) for removing the unwanted spin contamination intrinsic to the openshell UHF wavefunction. As we shall see when computed properties are ultimately the goal, it is far easier to work with the densities than it is to work with the wavefunction. The
PAGE 56
47 method we detail below is then centered on how the densities are affected as the spin contamination is diminished or removed. It was not our intention to rederive the algebra for projecting out the 5 = component of the spin density matrix. We instead give an overview and list the pertinent equations and citations. When expressed in terms of the corresponding orbitals the UHF wavefunction can be written in a form similar to that of a Configuration Interaction wavefunction [56, 57], e.g. equation (3.15), or where the Cs are functions of the corresponding orbital overlap integrals. The ^ are linear combinations of Slater determinants constructed from the corresponding orbitals, and are eigenfunctions of S\ The excited terms, Â• Â• Â•, will contain the 5^ = (p q) componentof the pure spin states 1^1 = ^{p Â— q), Â— 9 + (p? + e/), where e/ is the excitation level. It is these higher multiplicity states, l^l > ^{pq), that will be removed upon projection. Relations for the Cg were given by Sasaki and Ohno [58, 59] and are reproduced here since they were used in analyzing UHF wavefunctions in this work. In equation (2.66) below, ujs is the "weight" of the components with multiplicity (2s + 1), or, the fraction of the wavefunction composed from the {2s + 1) components, S Us = CI = (26 + l)^^i^ ^(l)^i?,B3^.+,(2.66)
PAGE 57
48 where s and are the quantum numbers associated with and Sz. While s is a summation index, = ^(p Â— 9). _ [(.m. + i)!p i!(2. + j + l)! ^ ^ (2) 1=1 and the T, are those of relation (2.36). If the wavefunction is normalized, then X;^s = l (2.69) To obtain an eigenfunction of a projector [41, 60, 61], Os, is applied to (2.24) or (2.65), where ds = l[At (2.70) and 01 = O s (2.71) S' Os = S{S + 1) ds is the normalized annihilation operator [41, 62] AS^^^(^1) .2 72) ^' ,(3 1) 1) (2.72) and t, of equations (2.70) and (2.72), ranges over all values, except t ^ s, consistent with the number of electrons in the system. The projection operator (2.70) projects out
PAGE 58
49 the {2s + 1) component of (2.65). The projected state wavefunction can be written as a linear combination of determinants [41]. O.^^H. = ^Ck{s,m,,^) T^^ (2.73) k=o The Ck{s,mz,^) [58, 59] are determined as r( n, , ,, (fm,fc)!(3 + m,)! C7,(.,mÂ„ ^) = (2. + 1) (2.74) Z.^ ^ /!(s_^^ + /_A;)!(f5/)!(26 + /+l)! The T^' are a sum of Slater determinants constructed from the corresponding orbitals [63, 64], note that (2.75) is not normalized. Tr = I'/r Â• Â• Â• ri;, v'^rn:,M''/3Vl3''] (2.75) The [a^~*^*^ is the sum of all spinfunction products, of the p Â— a(u;) and k j3{uj) spinfunctions, e.g. [a'' I = aa/? 1a^a + jSaa (2.76) and a*^'~*^] is similarly defined. Then [aP~*^*a''^'~^] is a sum of terms each containing p a{uj) factors and q /3{u)) factors [41, 65]. Harriman [66] showed that the above described projection leaves the eigenfunctions of the spin density matrix, as well as those of the firstorder reduced density matrix, (2.35),
PAGE 59
50 unchanged. Only the eigenvalues change upon projection and both sets, projected and unprojected, can be expressed as functions of the corresponding orbital overlap integrals, equation (2.36), and the {Cjt}, equation (2.74). In terms of the natural orbitals of the reduced firstorder density matrix, (or the charge density matrix) relations (2.392.41), the natural orbitals of the spin density matrix are expressed as \ <^f ^ (A. + /^.) , (2.78) Â•^f' = (A. (2.79) t = K (2.80) In eigenfunction form the relationships are expressed as roi = ikei ' (2.81) where ^ is defined as r = Y.ikei{i')ei\i) (2.82) and n = l 6 = (lr)^ k = D^\,q (2.83) ei^l', ik = {lTl)^, k^D + l,q (2.84) ei = %, 6 = 1, k = q + l,p (2.85)
PAGE 60
51 As was the case for the charge density natural orbitals, there are no relations for the unoccupied orbitals {E}. The Tk of the doubly occupied orbitals {D} all equal unity and thus have occupation numbers of zero, see equations (2.83) and (2.84) above, and so do not contribute to the spin density matrix. The validity of relations (2.782.80) can be checked by expanding them in terms of the corresponding orbitals and applying Â— Â— p~, where ^ = E^."(l')'7r(l) (2.86) r P^ = E'?f(l')'?r(l) (2.87) s and check against relations (2.832.85). If we now follow Harriman's [66] line of reasoning we can obtain the above results from a different perspective. We also will arrive at a point where Harriman's observations on the effects of projection are apparent. We now want to construct the operator matrix ^ in the basis of the charge density natural orbitals, and then obtain its eigenvalues and eigenfunctions. First we should note that can couple differing orbitals only if the differing orbitals are one of the pairs of "paired" orbitals, relations (2.39) and (2.40). Then the largest nonzero subblocks will at most be of order 2x2, and lie along the principle diagonal. This reduces the problem
PAGE 61
52 to diagonalizing a series of 2 x 2 matrices of the form f{K\r\^,) .2 88) to obtain the eigenvalues, and since (2.88) is selfadjoint it can be diagonalized via a unitary transform. If we use the Jacobi method [67] to obtain the eigenvectors we will gain an insight on the effects of the projection on these eigenvectors. We can use the unitary matrix , v /cosi9, sindi . n 0 , (2.89) sm Vi Â— cos Vi to diagonalize (2.88), and obtain its eigenvectors and eigenvalues. Once i?, is determined the eigenfunctions are obtained from cos(i9,) A,+ sin(i?,) fii (2.90) sin(i?,) A, Â— cos(i?,) Hi and i9, is determined from ^^2tf.Wm+WÂ£!M (291) now {Xi\r\\i)^{fii\r\f^i) = Q (2.92) and {^^\r\f^^) = {^^^\r\M)={lTf)^ (2.93)
PAGE 62
53 then matrix (2.88) is of the form f 0 (ir?)n v(i T^)' 0 ; (2.94) which has eigenvalues of Â±(ir2)i (2.95) From relations (2.91) and (2.92) we see that tan2??j = oo, or t?i = for all pairs {i, i'). This leads to ... being the eigenfunctions of (2.88), or, the same eigenvalues and eigenfunctions found previously in relations (2.782.79) and (2.832.84). Harriman solved the matrix elements of (2.91) for the projected spin density operator^ and obtained i9 = the same as for the unprojected density. Thus the {<^f } and {f } are unchanged by the projection, as are the singly occupied {'^}. (The interested reader is referred to References 62, 66 and 68 for a detailed discussion of spin projection on a UHF wavef unction.) We give Harriman's relations for the projected spin density matrix eigenvalues below. These are the relations used in this work to explore the effects of spin projection on the 6. Table IV of Reference 66. Harriman solved for the projected eigenvalues, and eigenfunctions, of p, p+ and p~ as well as for jo^. cosdi sini?,sin t?i Â— cos (2.96)
PAGE 63
54 calculated aisn(2.97) The relations (2.97) are for the special case of l^l = 5^ [66]. The ^, and 6' are paired eigenvalues that in the unprojected state sum to zero, while the have no pairing properties and in the unprojected state are equal to unity. The ujs{i) term is defined as gl ^s{i) = 5](l)^Cfc(5,mÂ„f 1) Afc(0 (2.98) A:=0 where (V) it Mi) = E fl^^D (2.99) and the Tj are those of equation (2.36). The uj^ term was defined in relation (2.66). It was not our intent to derive the relations of equation (2.97), to do so would fill half a Dissertation with unoriginal work. Harriman and others [41, 51, 56, 57, 6062, 66, 6873], and references within, have worked through the tedious algebra and arrived at (2.97). We have implemented (2.97) for the expressed purpose of investigating the effect of spin projection on the a.^o calculated for the large molecular systems we are interested in describing.
PAGE 64
55 The AUHF and EUHF Methods There are two methods related to PUHF that we have not mentioned, and, except for the following example, will not pursue. These are Annihilated UHF (AUHF) [74, 75] and Extended UHF (EUHF) [70, 76]. In Annihilated UHF one or more of the higher multiplicity components are removed from the wavefunction tfirough the application of the annihilator Ag [41]. As in PUHF the As is/are applied after the SCF, and thus the annihilated wavefunction is no longer stationary on the UHF PE surface. In practice it is usually the 25"^ + 2 component that is removed, and it is hoped that the higher multiplicity components lie too high in energy to contribute significantly to the spin contamination [43]. We have applied the Ag+i, s Sz, to the INDO/lUHF wavefunction describing the Bacteriochlorophylla (Bchla) and Bacteriochlorophyll6 (Bchlfe) cations [77], with the following results for (S^) and the aN of the four pyrrole nitrogens. Table 21: The affects of annihilation on {S") and calculated a.so. Bchla+ Bchl6+ Experiment^ UHF AUHF 1 Experiment UHF AUHF 0.750 3.970 13.36 0.750 4.268 14.79 0.78 2.24 4.68 0.82 1.80 8.75 aN2 1.12 18.4 38.3 1.17 16.2 78.4 aiVa 0.83 12.1 25.4 0.82 10.6 51.6 aN, 1.03 2.73 5.52 1.06 2.43 11.5 7. From Reference 77. 8. The units are Gauss, and the aN are for ^*N.
PAGE 65
56 From Table (21) we can see the assumption that the higher multiplicities contribute littie to tiie total wavefunction is wrong. As {S^) increased from 3.970 to 13.36 upon removal of the 5 =  component of tiie INDO/lUHF wavefunction for Bchla+, and from 4.268 to 14.79 for Bchlfe"^. Furtiier we see tiiat the INDO/lUHF calculated a.^o are all too large, and more than double upon annihilation. It is clear that the components with multiplicities greater than {23^ + 2) are contributing significantly to '^uhf and the computed a,so reflect this. (We should note that for small molecular systems annihilation does improve tiie {S^) as well as tiie calculated energy [78, 79].) Then from Table (21) AUHF is fatally flawed, at least when viewed from the perspective of wanting a method applicable to large molecular systems. The Extended UHF method mentioned previously is basically a spinprojected unrestricted fvmction where the orbitals are chosen to minimize the energy after projection. The EUHF wavefunction is stationary with respect to its variational parameters, and the spin densities calculated with it tend to lie between those determined via UHF and PUHF [70]. The EUHF methods are computationally cumbersome, and as stated above the resultant spin densities are no better the those of PUHF, at least in the sense of removing the unwanted spin contamination. For these reasons the EUHF method was rejected as a possible tool for exploring spin distributions within large molecular systems.
PAGE 66
57 Calculated Results for the UHF and PUHF Methods. In Table (D8) we present the results for calculating the aH and aN of several more test structures using the parameters of Appendix D, and the unpaired spin densities produced via the methods discussed in this chapter. In Appendix D we found that the INDO/1 Hamiltonian yielded consistently worse results, compared to those of the INDO/S Hamiltonian.^ For this reason we chose to limit our presentation to the INDO/S results. Along with the additional test structures we also present results for Bchla and Bchl6 cations and anions, as well as the Bphea anion. These are the types of compounds we want to be able to describe properly. The structures were obtained from the Brook Haven National Laboratories crystallographic data bank, through a collaboration with Dr. Jack Fajer, their number system can be found in Figure (15). We note that the hydrogens of the three crystal structures were not determined experimentally but were placed in position via a best guess algorithm. Upon inspection of the Bchl structures we found several "strange" methyl groups and repositioned the hydrogens associated with them in a manner consistent with chemical common sense. All calculated results will be presented in units of Gauss (G). 9. See Tables (D3) through (D5).
PAGE 67
58 Table 22: Calculated results for Bchla. Magnetic Calculated aÂ„o System Center^o Position^ ^ UHF PUHF Bchla+ 2.11 62 N 21 0.35 0.37 0.78 N 22 6.10 6.02 1.12 IN ZD 2.88 2.86 N 24 1.61 1.58 1.03 H 5 5.84 6.82 0.84 H 10 7.30 8.54 0.46 H 132 3.23 2.07 0.58 H 20 1.55 1.74 0.46 Bchla2.56 55 N 21 2.05 2.03 Â±0.42 N 22 1.42 1.43 2.32 N 23 0.38 0.40 Â±0.19 N 24 1.76 1.72 2.09 H 5 6.01 7.16 3.43 H 10 7.19 8.42 2.46 H 132 2.98 1.64 2.22 H 20 2.12 2.47 <Â±0.18 10. The isotopes are "iV and ^H. 11. See Figure (15). 12. Reference 77.
PAGE 68
Table 23: Calculated results for Bchl6 and Bphea. Magnetic Calculated Ui.^ System Center Position^ ^ UHF PUHF a Bchl6+ 2.44 59 N 21 0.74 0.75 0.82 N 22 9.43 9.23 1.17 IN Zj 2.24 2.24 0.82 N 24 1.51 1.49 l.UO H 5 7.42 8.28 0.89 H 10 8.64 9.59 0.43 H 132 2.25 1.76 0.43 H 20 1.03 1.10 0.43 Bphea2.47 58 N 21 0.58 0.55 0.43 N 22 7.15 7.07 2.56 N 23 0.78 0.75 0.21 N 24 1.83 1.71 2.19 CH3 2' 3.75 2.71 2.53 H 5 8.82 9.59 3.02 H 8 5.31 4.17 0.65 H 10 8.10 8.97 2.85* CH3 12' 4.67 5.30 2.96* H 20 5.80 6.58 2.47 NH 21' 0.89 0.84 0.35 NH 23' 1.10 1.07 0.17 13. See Figure (15). 14. Reference 77.
PAGE 69
60 The results presented in Tables (22) and (23) are somewhat disappointing. Neither method reproduces the experimental observations very well. All four cases were rather badly spin contaminated, as implied by the value of (5^). While the PUHF method is an exact eigenfunction of 5^, the results of this method still paralleled those of the UHF method. The only apparent affect of the projection was to diminish the \p'^\ by approximately , as is expected by the form of equations (2.97). Obviously, after projection, the of the two methods stayed proportional to one another. This is surprising, and disappointing as we had hoped that the projection would yield values in better agreement with experiment, having removed all spin contamination from a highly contaminated UHF wavefunction. Neither the signs of the splittings nor the their magnitudes were properly described by either method. At least not in a consistent manner. We should also note that the * values are considered to be suspect by those who took the measurement. It is obvious from these results that neither the UHF or the PUHF methods are capable of properly describing the unpaired spin distribution within these molecular systems.
PAGE 70
CHAPTER 3 THE ROHFCI APPROACH Introduction In this chapter we shall treat the openshell electrons differently at the SCF level than we did in Chapter 2. In Chapter 2 we used Pople's DODS approach and obtained a single determinate wave function which was an eigenfunction of S'z, as it must be, but not an eigenfunction of 5^ In this Chapter we shall use an SCF procedure that produces a wave function that is an eigenfunction of both these operators. Again as in Chapter 2 it will help if we outiine the details of this method, the Restricted OpenShell HartreeFock (ROHF) method, as it has been developed and implemented by Zemer and Edwards [80]. Restricted OpenShell HartreeFock Restricted OpenShell HartreeFock was first proposed by Roothaan in 1960 [81]. Since that time numerous methods have been proposed to address the problem of openshell molecules at the SCF level of theory. A sampling of these methods can be found in References [8192]. We outline the ZemerEdwards method below. The total wave function is constructed from a linear combination of Slater determinates. The coefficients of the Slater determinates are chosen to represent a particular spin state [93], to guarantee that this is an eigenfunction of 5^. The coefficients are 61
PAGE 71
62 determined by spin symmetry. Focklike operators are constructed for each shell, i.e. closedshell, 1st openshell, 2nd openshell etc. The constraint (<^.F''F''<^,)=0 (3.1) is enforced and guarantees the matrix of Lagrange multipliers is Hermitian, where Fi" is the Focklike operator for the /^th shell and F" for the i/th shell. The Hermiticity of each Focklike operator guarantees an orthogonal set of molecular orbitals for each operator. Though the various sets of molecular orbitals, one set for each openshell, will not, in general, be orthogonal to one another, as each has been constructed from the diagonalization of its own operator F**, or i but in general (<^ri<^;) = (^ri<)7^o (3.3) A concatenated set of orthonormal molecular orbitals is guaranteed in this procedure by projecting out the eigenvectors of each successive Focklike operator in the orthogonal complement of all those proceeding it. The spin density, equation (1.13), determined from an ROHF wave function is everywhere positive, a consequence of the construction of the ROHF wave function. Although negative spin densities are generally small, this prediction is clearly wrong.
PAGE 72
63 Direct experimental observation shows the sign of the unpaired spin density can be negative as well as positive. We have lost some of the flexibility we had with the UHF wave function, a price we paid to obtain a simultaneous eigenfunction of the Sz and 5^ operators at the SCF level. A single spinadapted configuration state function lacks the flexibility to describe differences in unpaired spin distributions. We would have to construct a total wave function of many such "spinadapted configurations" to have the degree of flexibility necessary for a proper, or realistic description. The simplest way to construct such a total wave function is through the method of Configuration Interaction The Configuration Interaction method is a straight forward application of the Ritz [94] method of linear variations as applied to the determination of electronic wave functions. Again we desire to solve the Schrodinger equation, equation (1.14), and again we must approximate its solution. In the CI approach the approximate wave function is expanded as a linear combination of "configuration state functions" (CSF) in an analogous manner as the LC AOMO approach, equation (1.42), for the HartreeFock molecular orbitals. (CI). Configuration Interaction N (3.4) s
PAGE 73
64 Where the {$} are the configuration state functions and the expansion coefficients, Cs, are the parameters which are varied to make E[^aj,^] stationary. N is the number of expansion functions. We examine, = / (*<.p.*<.p.) (3.5) in which H in equation (3.5) above is the electronic Hamiltonian, equation (1.20). This leads to the general matrix eigenvalue equation HC = SCe (3.6) where Hst = {^s\mi) (3.7) and = (3.8) The linearly independent eigenvectors Cp, the columns of C, can be chosen to be orthonormal with respect to S, N = S ^tp^^tc., = (3.9) s,t leading to the simple eigenvalue equation Hcp = c^E, (3.10)
PAGE 74
65 If we label the corresponding eigenvalues such that Ei
PAGE 75
66 rearrangement of electrons within partially filled shells. The CI method recovers the correlation energy, both dynamic and nondynamic, at least in the limit of a full CI. (We will explain the various CI expansions later.) While the inclusion of correlation into an abinitio wave function is necessary for the accurate description of observable properties, it is a less than good idea for semiemperical methods. The reason this is so is straight forward; semiemperical methods are parameterized. These parameters include ionization potentials and electron affinities as well as completely empirical parameters such as P^,^, equation (1.82), which are varied so as to reproduce some benchmark. If the parameterization is based upon experimental values then correlation is included to an unknown degree. An example of how correlation is built in can be seen in the MatagaNishimoto formulation [97] of the twocenter twoelectron integral, equation (1.80) where Jaa = IPa + EA, (3.14) the sum of the ionization potential and the electron affinity. The experimentally measured IP and EA are already "correlated"; any and all such effects or interactions are already embodied in these measured quantities. To now introduce correlation into the INDOSCF wave function is to, in an indeterminate way, "doublecount" the correlation. The amount
PAGE 76
67 of correlation that is built in depends on the level of theory that is parameterized and what parameters are chosen. Single Excitation Configuration Interaction We now need to address the configuration interaction method used in this work. The CI basis, {$}, is constructed from a set of spin molecular orbitals obtained from a INDOROHFSCF calculation. The expansion functions $ are linear combinations of Slater determinates whose coefficients are chosen such that the CSF's are eigenfunctions of botii Sz and S\ where (5^) and (S^) of {$} are tiie same as tiiat of the INDOHF reference function. Such configuration functions are said to be "spinadapted". The method used to construct these spinadapted functions and the algebra imposed by their form in the evaluation of the spin density will be addressed in detail later. We now need to examine the expanded wave fimction in more detail. We can write equation (3.4) in a less general and more insightful form as occ vir occ vir occ vir = co$o + E E + E E + E E ^^k^^H + Â• Â• Â• (^.is) i 1 i,j a,b i,j,k a,b,c where $o in the first term of equation (3.15) is the ROHF reference. The set {$^} of the second term is comprised of all spinadapted configurations generated by removing one electron from an occupied spin molecular orbital ^, and placing it in the virtual, or unoccupied, spin orbital tpaThese are known collectively as the single excitations. The is generated by removing two electrons from the occupied molecular orbitals, or
PAGE 77
68 occupied space, and placing them in the virtual space. Such spinadapted configuration functions are also known as double excitations or doubles. The same pattern holds for the triple and higher excitations. If the summation of the various excitation levels in equation (3.15) is carried to its limit, then the calculation so performed is said to be a full CI calculation. The number of configuration functions A'^ used in the construction of a full CI wave function, for a given number of electrons n, orbitals M and total electronic spin angular momentum S is given by Weyl's formula [98] which also equals the dimension of the full CI Hamiltonian matrix. Obviously full CI calculations are, from a computational point of view, intractable for all but the smallest of molecular systems. For this reason the CI expansion, equation (3.15), is most often truncated to include at most only the first two or three terms. Even with this reduction in the size of the problem it is not unusual to reduce it even further by limiting the size of the "active space". That is to limit the number of molecular orbitals from which electrons are removed from or placed into, usually centered around the highest energy occupied molecular orbital (HOMO) and the lowest energy unoccupied molecular orbital (LUMO), or the "HOMOLUMO gap". For the work in tiiis thesis the CI expansion, equation (3.15), is truncated at the second term and all single excitations, are included in the CI wave function, and it is referred to as a "singles excitation CI" or
PAGE 78
69 a CIS calculation. We can now come back to the problem of the "doublecounting" of correlation. If we construct a simple trial wave function comprised of the closedsheU SCF reference, $0, and one singly excited configuration function, or *c/s = (Co$o + Cf^n (317) and proceed to evaluate {^cis\H^,\'^cis) for its lowest eigenvalue and associated eigenvector we obtain the matrix eigenvalue problem It is obvious that any mixing of the diagonal elements is through the offdiagonal element ($o^e,$n = ma) + [(^'H"^) (^H^a)] (319) r or its adjoint The mth element of the Fock operator, again in the basis of spin molecular orbitals, is given by mm = {i\h\a) + Yl^{ir\ar) {ir\ra)] (3.20) r Notice that the righthandside of both equation (3.19) and equation (3.20) are identical and thus {^oA,m = {Hm) (3.21)
PAGE 79
^ 70 Now by definition solving the HartreeFock eigenvalue problem requires the offdiagonal elements of the Fock matrix to vanish, or upon solution ' Â• ^ (^.im) = o :; ' (3.22) and then so must equation (3.19). The above exercise is an example of Bnllouin's Theorem [99]. Which is stated in words as; singly excited Slater determinates will not directly interact with a closedshell HartreeFock reference determinate $o) through the electronic Hamiltonian, or ($o^.,$n=0 (3.23) as well as its adjoint. Restricting the CI expansion to the {$"} leaves the closedshell HartreeFock reference uncorrelated, and so the problem of doublecounting correlation has been avoided. This approach to the CI method has proven quite successful for the calculation of the low energy ultraviolet, and visible, absorption spectra of molecular systems [100103]. This success was the primary reason we chose this technique as a starting point for the calculation of (jo^(r;^)). Unfortunately, at least from the point of view of doublecounting correlation, equation (3.23) does not hold for an openshell HartreeFock reference. Certain single excitations from an ROHF reference fimction lead to spinadapted configuration functions which differ from the reference by two spin molecular orbitals and such configuration functions
PAGE 80
71 interact directly with $o), and in general ($3^.,$^/)^0 (3.24) An example can be given by letting the wave function equation (3.37) interact with the wave function equation (3.41) through the electronic Hamiltonian, vis {T,\H^,\T,) = ^{im\ma) (3.25) V6 It is this direct mixing of the reference function with "certain" singly excited functions, and through them indirectly with the other single excitations, that gives the CIS approximation to the molecular ground state the flexibility to describe differences in the alpha and beta spin distributions; unlike the ROHF reference function. This leads to the question; can a method that was carefully crafted to avoid such interactions as equation (3.25) above now be expected to well describe another property, and can the introduction of correlation to an already "somewhat" correlated reference function be done in such a way as to obtain consistent results? The first question above is answered by this thesis and the second is answered in Table (31) below.
PAGE 81
72 Table 31: How the correlation of the INDOROHF reference affects the calculated value of a.jo. % Correlation System aH ExD^ A flH CIS CISD^ ^OH 26.4 3.34 8.31 100.0 23.0 2.49 7.47 98.4 20.2 1.98 4.23 98.6 2BH315.1 2.05 5.50 98.4 25.1 3.53 17.7 100.0 2NH3+ 25.9 3.10 9.99 96.0 2hcn137.1 4.14 3.92 93.5 202H 9.8 1.35 1.38 96.7 In Table (31) the (aH Exp) column lists the experimental Hydrogen aiso values, and (A aH) represents the difference in the Hydrogen's calculated isotropic hyperfine splitting constants when calculated from a CIS and a CISD wavefunction. The (% Correlation) columns lists the fraction of the correlation energy recovered, in percent, when compared to a full CI. It is of interest to note the average (A aH) is 2.75 and its standard deviation is 0.93; recall that a,so varies almost linearly with respect to p^, equation (1.10). A comparison of columns (aH Exp) and (A aH) indicates that regardless of the magnitude of aiso the shift is a rather constant 3 Gauss. Inspection of the CIS and CISD (% Correlation) columns shows that the CIS wave function has recovered a small fraction 1. All aiso values are reported in units of Gauss. 2. CISD calculations include the first two terms of equation (3.15) in the construction of the approximate wave function.
PAGE 82
73 of the correlation energy, which is expected due to the generalized Brillouin theorem [104] while the CISD wave function has recovered almost all of it. Nevertheless, the calculated property a,so is for all intents and purposes the same for either wave function. This demonstrates that the calculated CIS a.^o values increase in magnitude by a small and almost constant amount when the wave fimction is fully or almost fully correlated. This small "constant" shift indicates that a CIS wave fimction has the same flexibility to describe the spin distribution as a more highly correlated wave function at least within the INDO/S model. This to a large extent relieves our concern with the possible problem of correlating an already correlated reference, and the possible inconsistencies that may have been introduced into our calculated a.soWe chose Hydrogen for the above endeavor for the following two reasons; 1) there is no core orbital spin polarization problem when calculating a,so using a valence orbital only theory and, 2) minimum basis set theories have problems describing the "breathing" of the atomic orbitals as a function of the atomic charge. By taking the difference in the computed aiso the magnitude of the contact has to a very large extent been removed, and the difference represents almost purely the difference in the spin density as computed by the two wave functions. This leaves (A aH) a good gauge of the effect of correlation on the calculated a,ao, even when using a minimum basis valence orbital only theory.
PAGE 83
74 SpinAdapted Configuration Functions There are many ways to construct simultaneous eigenfunctions^ of Sz and 5^ when starting from a set of HartreeFock spin molecular orbitals {V*. }Each method entails a unique procedure for the evaluation of the matrix elements of the CI Hamiltonian, a subject we shall not be concerned with, as well as for any integrals necessary to evaluate the expectation values of observable properties, a subject that shall concern us. In this work two Configuration Interaction procedures were used in the evaluation of a,ao. One is quite general and in principle can treat molecular systems of any multiplicity and of any openshell structure; it is also quite slow. The second CIS procedure is specific to a singleÂ— Â« 1 reference, singlesexcitation, calculation on molecular systems with [Sz) equal to , or doublet molecular systems; it is also quite fast. Since at the time of their implementation the procedures for evaluating (^(r^)} for either CIS method were unique, we shall present them in some detail here. Rumer Bonded Functions A generalized extended Rumer diagram methodology [105107] has been presented by Cooper and McWeeny [108, 109]. Their formulation was general to any matrix element between bonded functions [106, 107] and included formulae for spindependent operators. We, while not being cognate of this work, also worked out formulae for matrix 3. An outstanding reference on the subject of "Spin Eigenfimctions" can be found in the book by Pauncz, Reference 93.
PAGE 84
75 elements of a spindependent operator, built upon the initial work of Zemer and Manne [110]. Since it was this algebra that was implemented we shall develop it below and not that of Cooper and McWeeny, through as one would expect they are similar. The valencebond type spin functions grew out of the work of Heitler and Rumer [105107] and of Slater [111], and are commonly called Rumer or Slater bonded functions.'* The configuration functions constructed to be spin eigenfunctions via the Rumer method are linearly independent [93], but not in general orthogonal. The Rumer functions constructed from the same set of spatial orbitals will not be orthogonal among themselves but wiU be orthogonal to all other Rumer functions, assuming an orthonormal set of spatial orbitals form the initial orbital product. The spin coupling in a particular configuration function based on the orbital product M.n (3.26) is described by a system of parentheses (^1 i2 (^3 Â• Â• Â• 0Â„.)^^Â„i)^Â„) (3.27) singlet spin coupling pairs of orbitals in the product. For a given number of electrons A'^ and a total electronic spin angular momentum S, there will be y 5 pairs of spin 4. We shall use the notation of Boys, Reeves and Sutcliffe [112, 113].
PAGE 85
76 coupled orbitals for the case of 5 = nis.^ For i / j {Mi) = M,{vMA^2)^H^rWM ^MccM] (3.28) for i Â— j (M,) = U.irMr.nc'MPM (3.29) The unpaired orbitals are represented by "dangling" left parentheses, again for i / ; {icf>j){>. = A<^.(rJ^,(r3)<^,(r3)^[a(ti;,)/3(a;,)a(ii;3) ^{u;,)a{u,)a{w,)] (3.30) During orbital rearrangements to obtain maximum orbital coincidence between two spinadapted configuration functions each orbital keeps its associated parentheses. Diagrammatically the Rumer spin functions are constructed in the following manner; the spatial orbital labels are placed at the vertices of a polygon with one extra vertex added as the "pole". Starting at the pole and proceeding around the polygon in a counter clockwise manner the orbital labels will be in ascending order. The labels are then connected by arrows into singlet coupled pairs, equation (3.28), with the arrow's tail at the lower index and its head at the higher index. As many such couplings as necessary are formed to obtain the desired Sz value. All uncoupled orbitals are cormected to the pole by 5. For a given N electrons and S electronic spin angular momentum there will be /iV'\ ? N' \ f{N',S)= I, j spin functions [114] for a given orbital product. Where N' equals the number of singly occupied orbitals and q' equals ^ S.
PAGE 86
77 arrows. Within the above rules all such possible diagrams are constructed and any with crossed arrows are rejected. Those with crossed arrows are not linearly independent [115] and can be constructed as a linear combination of those diagrams whose arrows are not crossed. For the case of a three electron three orbital S = ^ system see Figure (2) below. Equation (3.31) translates the diagrams of Figure (2) back into the conventional algebra 1 1 1 2 3 Dl 2 3 D2 Figure 2: Rumer bond diagrams for the case of three electrons in three orbitals and S Â— ^. Dl = {(f>i(l>i){3 ^\(f>i(l)24>i\{oil3a I3aa) D2 = {M.){(f>, = ^\MsMa/3a ^aa) (3.31) where the electron labels are assumed to be in canonical order and have been suppressed. By inspection it is obvious the Dl + Z)2 = D'i. Armed with the formalism of this section we are ready to describe the evaluation of {p^{vn)) using a Rumer CI wave fimction.
PAGE 87
78 Evaluation of for a Rumer CI Wavefunction The evaluation of matrix elements betvi'een Rumer functions for the spin density operator, p^, is achieved as follows. Place the Rumer functions in maximum coincidence, then connect each spin molecular orbital in the first function to its counterpart in the second with a vertical bar. vis {MM'l>.){. = (1 2) (5 (3 4) I I I I I (3.32) {M2)iM. = {l 2) (5 3)(4 Then horizontally connecting the coupled pairs and suppressing the indices leaves a "closed loop" and a "odd chain", vis Figure 3: A closed loop and an odd chain. For cases where S is greater then ^ "even chains" are also possible, vis Figure 4: Two types of even chains which arise from (12) interacting with (1 (2 in the first case and (1 2) (3 4) interacting with (1 (2 3) (4 in the second. and in general so are multicycle closed loops, vis
PAGE 88
79 Figure 5: A multicycle closed loop, arising from the interaction between (1 2) (3 4) and (1 (2 3) 4). Once the matrix element between a pair of bonded functions has been reduced to the above structures, the evaluation of (p^) is straight forward. The operator will couple one orbital pair, linked by a vertical bar, at a time and {p^) is then just the sum of these couplings. Now from the properties of the closed loops their total contribution to {p^) sums to zero. This is a direct consequence of {,^,) = {,ct>,) (3.33) If an even chain is formed then the contribution of the matrix element between the two Rumer fimctions is zero. There will always be spin orthogonal orbitals coupled in such a case, and no oneelectron operator leaving the primitive spin functions unchanged, such as p^, can couple even chains. This leaves only the odd chains contributing to {p^{rN)). Thus only the density contributions of the orbitals coupled by vertical bars in the odd chains need be computed, and (^^(r;^)) as evaluated from a Rumer CI wave function
PAGE 89
80 is written as (3.34) I, J Is or in the atomic orbital basis {nr.)) = E "^^C.r,, Y.l'H nAxMv r,)x.) (3.35) i,j L s where the summation index (5) runs over all the vertically linked orbital pairs, 9* is a link parity factor and alternates as H h Â• Â• Â• +, the summation indices // and v run over all atomic orbitals, the coefficients Cj and Cj are the CI expansion coefficients, see equation (3.4), for the /th and Jth Rumer functions respectively and the indices / and J run over all such bonded functions. The atomic orbital density P^^ is determined from the 5th vertically linked molecular orbital pair. The factor Tjj accounts for the product of the normalization factors of the /th and Jth bonded functions as well as phase changes due to orbital permutations within the bonded functions necessary to obtain maximum coincidence between them. In equation (3.36) (/) is the number of closed loops, (6) is the number of bonded pairs (k) is the number of bonded pairs with identical orbitals in one function not matched by a bonded pair of identical orbitals in the other, the factors (Ti and aj are the number of permutations necessary to bring the unpaired orbitals back to their original order in bonded functions / and J respectively. T,, = {2)'\2)Hl) (3.36)
PAGE 90
81 Fast Doublet CIS Whereas the Rumer bonded function method for the construction of spinadapted configurations was very general with respect to the multiplicity, number of openshells and degeneracies within the openshells, the following method for their construction is very specific, and entails a new formalism for the evaluation of (p^(r;v)). For lack of a better name we shall call this method the "Fast Doublet CIS" (FDCIS) method. The FDCIS algorithm is specific to the following case, only singles excitations off a single reference determinate of spin S Â— ^. This restriction allows for concise and easily evaluated matrix elements of the CI Hamiltonian [116], again this topic lies outside the scope of this work. We shall detail below the formulae for the evaluation of (p^(rjv)) from a FDCIS wave function. FDCIS SpinAdapted Configurations Within the FDCIS formalism there are four unique molecular orbital configurations, see Figure (6), these lead to five "types" of spinadapted configuration functions. We shall label these generically as To Â• r4 and shall define them below. In an attempt to achieve
PAGE 91
82 in To Ti T2 T3&4 Figure 6: The four possible orbital configurations for the FDCIS formalism. compactness of form the following convention shall be followed; a Slater determinant, l^i^i^^ ^n\, will be represented as \iijn\. The orbital indices i,j,a and m shall take on the following meaning, i and j will represent doubly occupied orbitals, (a) shall represent a virtual orbital and m shall represent the openshell orbital in the reference function. Any subscripts on the orbital indices will indicate the "type" of configuration function from which it came, it will be clear below what this means. Now let us write the five types of spinadapted configuration functions To = '^o = \llm\ (3.37) T, ='$7 = 11 Â•Â• mmT, = '^l = \n.a\ T, = = iam\ + \llaim\) (3.38) (3.39) (3.40)
PAGE 92
83 = = Â• ia Â• Â• Â• m 11 Â• Â• Â• ia Â• Â• Â• m + 11 Â• Â• Â• Â• Â• Â• m) (3.41) where the superscripts and subscripts on $ indicates the single excitation, except for To which is the ROHFSCF reference configuration function. Formulae for the Matrix Elements of Pf^cis Again in the interest of compactness we shall express the matrix elements of the various types of spinadapted configurations, coupled through p^, as the sums of the orbital densities they contribute to p^^. Let us start with the diagonal elements and recognize we are working with real spatial molecular orbitals, or Â— {To\r\To) = \m){m\ (3.42) {TAr\Z) = \j){j\ (3.43) (r.l^^ir,) = \a){a\ (3.44) (r3p^r3) = \m){m\ (3.45) {T.\r\T.) = f + \a){a\ im)(m) (3.46) now the offdiagonal elements of the type {Tj\p^\Tj) where Tj ^ T'j {TAr\T[) = mj'\ (3.47) {T.\r\Tl) = \a){a'\ (3.48) (73 1^17;')= 0 (3.49)
PAGE 93
84 {T.\r\T:) = i\iW\ ifa = a' (3.50) = a)(a' if i i' now the offdiagonal elements of the type {T^\p^\TÂ„) Irri 1 1/77 \ = biKm (3.51) {To\r%) (3.52) {To\r%) = 0 (3.53) {To\r\T.) (3.54) = 0 (3.55) {TAr\T.) if ii = Â«3 (3.56) {T,\r\T.) if ii = Â«4 (3.57) {nr\T.) = ^V2){m\ (3.58) {TAr\T.) if z'a and 03 = 04 (3.59) = ^a3)(a4 if ^3 = ^4 and 03 04 (3.60) = 7^(a)(a if = and Ga = 04 This leads to {r{r.)) = C*^^ E Kt'^xMv v,)\x.) (3.61)
PAGE 94
85 where / and J run over all configuration functions. The Ci and Cj are tiie CI expansion coefficients, equation (3.4), tiie indices /x and v run over all atomic basis functions, {x}. and the P'J are obtained from equations (3.42) through (3.60). Calculated Results for tiie ROHFCIS Method As in Chapter 2 we again present results for several smaller test systems similar to those used in determirung the optimum valence Slater sorbital exponent, in Table (D8). As was the case in Chapter 2, the INDO/1 Hamiltonian yielded consistentiy worse results than did the INDO/S Hamiltonian when compared to experiment.^ For this reason we limit our presentation of results to those of the INDO/S Hamiltonian. The Bacteriochlorophyll structures and the Bacteriopheophytin structures are those used in Chapter 2, and again all calculated results are presented in units of Gauss (G). 6. See Tables (D3) through (05).
PAGE 95
86 Table 32: Calculated results for Bchla and Bphea. Magnetic Calculated System Center^ Position^ Bchla + N 21 0.09 0.78 N 22 0.47 1.12 N 23 0.52 0.83 N 24 0.08 1.03 H 5 0.35 0.84 H 10 1.46 0.46 H 132 , 1.36 0.58 H 20 0.08 0.46 Bphea N 21 0.62 0.43 N 22 1.21 2.56 N 23 0.09 0.21 N 24 1.58 2.19 CH3 2' 2.47 2.53 H 5 3.28 3.02 H 8 0.57 0.65 H 10 0.83 2.85* CH3 12' 0.51 2.96* H 20 3.67 2.47 NH 21' 0.15 0.35 NH 23' 0.04 0.17 7. The isotopes are and ^H. 8. See Figure (15) 9. Reference 77.
PAGE 96
87 Table 33: The calculated results for BcMft. Magnetic Calculated System Center Position a,so a " Bchli + N 21 0.13 0.82 N 22 0.12 1.17 N 23 0.69 0.82 N 24 0.13 1.06 H 5 0.24 0.89 H 10 0.64 0.43 H 132 1.43 0.43 H 20 0.23 0.43 Comparing the aÂ„p and calculated aÂ„Â„ columns of Table (32) and (33) we immediately see that the ROHFCIS results are good. The computed sign of the hfc are all correct. Unlike the UHF and PUHF methods the computed magnitude of the hfc are all reasonable. If we bear in mind that the experimental hfc were obtained in solution, and the structures are from xray crystallographic methods, then these results do indeed seem reasonable. We should also note that the * values are considered to be suspect by those who took the measurement. While it required obtaining the eigenvector associated with the lowest eigenvalue of a matrix of order 26000, these results seem worth the effort. 10. See Figure (15) 11. Reference 77
PAGE 97
CHAPTER 4 CONCLUSIONS Several points need to be made before we can draw our final conclusions. Whether the tools we use to investigate nature are totally empirical, semiemperical or as close to first principals as the situation will allow, there is only one real measure of success; and that is, will our model describe observed reality? This leads to the following question, what is observed reality? Or stated in a less philosophical manner, how good are our measurements of nature? The qualification of this question will shade light on the "goodness" of our model. Our model treats a single molecular system as the sole inhabitant of its universe. It interacts with nothing but itself and its conformation is fixed. Compare this to the world the experimentalists' molecular systems occupy; most experimental observations are made in the solid of liquid phases. (All molecular systems used in this work as measures to obtain our parameters, or as yard sticks to measure the parameters' "goodness", were themselves measured in the liquid phase.) In the few cases where both gas phase and condensed phase ESR spectra are available for the same radical species, the hfc typically differ by a few percent, but can differ as much as 10+ % for the extreme cases. Computed properties such as spectroscopy, dipole moments, transitions moments, 88
PAGE 98
89 etc. as well as ESR hfcs are more sensitive to structural changes in the molecular system than is the computed energy of the system. The hfc are particularly sensitive to such changes [117, 118]. Bartlett et al [119] in a work to be published, show this quite convincingly for triatomics, at several levels of theory. The structures we used in the calibration of our model were not all determined experimentally, many were obtained via quantum chemical techniques. The conjugated systems were optimized via an INDO method, where practical experience has shown the resultant structure to be as good as, or better than, those obtained via abinitio SCF methods. The alkane structures were obtained using abinitio methods, as the INDO method we use has known deficiencies when applied to such systems. In either case it is physical intuition and low level quantum chemical methods by which the structures used in this work were obtained. Others who work with smaller systems use methods which are capable of producing more precise structures, usually gas phase structures; but all in this business must ask, "is the structure we are using the same as Uiat Uie experimentalist is measuring"? The bacteriochlorophyll and Bacteriopheophytin structures used in this work were obtained from the solid state, and the hfc , aÂ„p, are from solution experiments. The "floppy" substituents of the Bchla, Bchl6 and Bphea are prone to large displacements upon going from the liquid to the solid state, and visa versa. As noted above the hfc are very sensitive to changes in structure, so we chose magnetic nuclei that were substituents of the macrocycle or the conjugated
PAGE 99
90 rings to use as measures in our calculations. Here we assumed that these rigid regions would be the least perturbed upon phase transitions. Molecular vibrations effect the measured hfc and corrections for this are typically a few percent of the measured hfc value [118]. In extreme cases where large atomic displacements are possible, and especially when such displacements move the measured magnetic center out of a nodal plane into direct interaction with the orbital describing the unpaired electron, such as the case of the methyl radical, the vibrational averaging correction can be of the order of the magnitude of the observed hfc splittings. The differing slopes of Figure (10) may be partially due to such vibrations. Vibrational modes that would bring the hydrogen lying in the nodal plane, out of the plane and into direct interaction with the tt molecular orbital describing the unpaired spin, would increase the unpaired spin density on the hydrogen. Though the effect would tend to reverse the sign of the spin density at the hydrogen; as the induced spin polarization at the hydrogen, due to the unpaired electron in the p^^ orbital of the adjacent carbon, would have a negative sign. Direct interaction between the hydrogen and the tt electronic system of the molecule would place positive spin density at the hydrogen. Multicenter exchange, which is missing from the INDO Hamiltonian, would of course, also effect the slopes of Figure (10), and this is something we will investigate in the future. A comparison of the CIS cr and tt ^i, of Tables (D6) and (D7) indicates
PAGE 100
91 that, at least, for the systems included in the exponent optimization the slopes should be very close. Figures (1214) are plots of a,,Â„ vs a,,p, for the UHF, PUHF and the ROHFCIS methods. The slopes of these plots are determined via a linear least squares fit and should equal 1.0; the intercepts should equal 0.0. The UHF and ROHFCIS slopes are close to 1.0 (i.e. 1.01 and 1.04 respectively), while the intercepts are respectively, 0.99 and 0.02. While the slopes for both methods are good, the intercept of the UHF method in not as good as that of the ROHFCIS method. The PUHF slope and intercept are both bad, compared to the other methods, and have values of 0.89 and 1.39 respectively. Since all data points plotted in these figures are for systems that were used to optimize ^, these values are as good as the methods are capable of. From above, and as we shall see later the PUHF method is the least successful of the three, followed by UHF, and the ROHFCIS method will prove to be the best. From Figure (10) we see that the twoand threecenter terms are necessary. If we do not account for them we obtain two for hydrogen. One for the a systems and one for the TT systems, the iÂ„ 1.48 while 1.20. Accounting for the twoand threecenter terms leads to = 1.248, and from Figure (14) we see that the inclusion of the twoand threecenter terms and using = 1.25 does lead to good results for hydrogen. The aforementioned exponent being that of the ROHFCIS method, since both the UHF and
PAGE 101
92 PUHF methods proved to be unsatisfactory, see Appendix D. Other researchers have shown that the polarization of the atomic core orbitals can be important, as well as electron correlation [118, 120]. Our model has no core orbitals and depends on the core effects being directly proportional to the valence contribution to ai,Â„. This has been shown to be the case by Chipman [118] for the second period elements when the structure is close to its equilibrium geometry. All of this leads us to ask "just how close to the experimental values can we reasonably expect to come"? We obviously can do no better than the limits placed on aÂ„p by interactions with the environment. Shifts in aÂ„p due to vibrations are not accounted for in single point calculations. Primarily for these reasons the best of the high level quantum chemical methods can consistently yield results of no better than Â± 20% for radical molecular systems. It would then seem reasonable to assume that our model should on average target Â± 20%. The results of Chapter 2 are disappointing. We had expected the removal of the higher multiplicity components of the UHF wavefunction to improve the calculated hfc. Instead, the wavefunction is sufficiently poor even after spin projection that no improvement was gained. Neither the sign nor the magnitude of the calculated hfc splittings can be trusted when obtained by either of these methods. It may prove interesting to take the natural orbitals of the UHF reduced firstorder density matrix, which are unchanged
PAGE 102
93 upon projection, and use them and their projected occupation numbers as a basis for a configuration interaction calculation, or a perturbative calculation, to obtain the spin density. For neither wavefunction is useful at the SCF level for determining a.^,,. The results of the ROHFCIS method are heartening. The signs of the hfc are correct in all cases. The calculated magnitude of the splittings are reasonable. We say this even knowing that some of the absolute errors are large (of course the UHF or PUHF errors were far larger). The Bacteriochlorophyll structures are suspect, and their results are not as close to experimental observations as one would have liked. The Bacteriopheophytin structure is considered as good as one can obtain for such molecular systems, and the Bacteriopheophytin calculated results are noticeably better than those of Bacteriochlorophyll. Several points need to be made or reiterated concerning this point. First, all hfc are quite sensitive to structural changes in the molecular system, and there is little chance that the structures we use in these calculations were identical to those measured experimentally. Secondly, for most splittings, and for all negative splittings, we are dealing with calculating the differences in small, induced, asymmetries in spin distributions. When calculating small hfc, usually for centers that do not directly contact the unpaired spin, (e.g. Those magnetic nuclei that lie at the nodes of the orbital describing the unpaired spin.) the method must to a high degree be capable of properly describing both the alpha and beta spin distributions. For it is the difference in these
PAGE 103
94 two distributions that determines the observed hfc. Seen in this light, this method is at least qualitatively useful, and may well be useful as an aid to spectra interpretation, see the results for Bpheain Table (32), or in conformation determination. Finally we note that measurements on most large systems are done in solution or in condensed phases. We look forward to modeling this effect using the selfconsistent reaction field method [121], which experience has shown to have major effects on structure [122, 123], UVVisible spectroscopy [124], and even in the calculation of molecular polarizabilities and hyperpolarizabiUties [125]. Finally, it may prove insightful to "classically" give these molecules a temperature and investigate how vibrations, and torsions in the conjugated system affect the spin distribution. Such classical mechanical approaches are now possible to apply to systems such as these.
PAGE 104
APPENDIX A UNITARY ORBITAL ROTATIONS For any wavefunction constructed as an antisymmetrized product of canonical HartreeFock orbitals, (in general any orthogonal functions) the expectation values determined from such a wavefunction are invariant to unitary rotations of the canonical orbitals. To see this let us take an orthonormal set of spatial orbitals {(f)} and with them construct a square matrix ^M^) Â•Â•Â• ^n(l) A= i Â•Â•. i I (A.1) of order n x n such that kM^) Â•Â•Â• 4>n{ri), ^ = 1= det{K) (A.2) Let us also define a second matrix of order n x n such that = (A.3) Such a matrix is said to be unitary and when used in transforming the {'i = J2rUri (A.4) r will preserve the orthonormaUty in the transformed set. Then from relation (A.4) it is clear that the matrix which corresponds to A but constructed from the transformed orbitals is B = AU (A.5) 95
PAGE 105
96 Now det{B) = det{AV) = det{A)det({J) (A.6) which is, to within a normalization constant, equal to 1^') = rfei(U)*) (A.7) Exploring relation (A.3) leads to det{U^U) = det{U^)det{U) = det{V)* det(U) = \det({J)\^ = det{l) = 1 (A.8) or in general det{V) = e*^ (A.9) Then ^' differs from ^' by at most a phase factor, and if U is a real matrix this phase factor is just Â±1. It is a postulate of Quantum Mechanics that the expectation value of any observable property depends on thus in any observable way, ^' = ^.
PAGE 106
APPENDIX B THE MIRROR THEOREM Let A be an arbitrary matrix of order m x n, and let B be an arbitrary matrix of order n x m with m > n. The products AB and BA are quadratic matrices of order m X m and n x n respectively. Theorem: The nonvanishing eigenvalues of (AB) and (BA) are necessarily the same, and the classical canonical forms associated with multiple eigenvalues are identical [43, 53, 126]. The mirror theorem as stated above is a more general theorem than we need. We choose to demonstrate the theorem for the special case of product matrices that are selfadjoint, since this will be the case throughout this work. A more rigorous proof of the general theorem can be found in the noted references. If S is Hermitian (i.e. = S) then a unitary matrix U (i.e. U+ = U"^) can always be found, where U^SU = A (B.l) and A is a diagonal matrix. For simplicity we assume BA to be Hermitian and have exactly n nonvanishing eigenvalues. Then there exists a matrix V of order n x n that 97
PAGE 107
98 brings BA to diagonal form V^BAV = A (B.2) Now consider U = AVAi (B.3) where we choose to use the positive square root of the A^.^ Then (AB)U = A(BAV)A^ = A(VA) A^ = (AVA>)A (B.4) = UA or U^(AB)U = A (B.5) Conversely we can start with the nonvanishing eigenvalues of AB, and show that, if V^ABV = A (B.6) then we define U = BVA^ (B.7) and (BA)U = B(ABV)A^ = B(VA) A"^ = (BVA^)A ^ (B.8) = UA 1. This forces the overlap between the "paired" orbitals, Appendix C, to be positive.
PAGE 108
Equations (B.2) and (B.5), as well as equations (B.6) and (B.8), are eigenvalue equations with the same nonvanishing eigenvalues, A. The mn remaining eigencolumns of AB could be repeat eigencolumns or could be zeros. Since both AB and BA are Hermitian and both have a rank of n, each then have n orthogonal eigencolumns, and the m n extra columns of AB are then zeros and so have zero length.
PAGE 109
APPENDIX C THE PAIRING THEOREM The pairing theorem is a special case of the muTor theorem,^ to see this and to demonstrate the theorem, let us consider two sets of basis fimctions, a^ia^aa, ,aÂ„,} (C.l) b = {b,,b, ,bÂ„} (C.2) of order m and n respectively, with m > n. Further we assume the sets to be orthonormal. Then one has the overlap integral matrices S^,Â„ = (ab) Â• (C.3) St_ = (ba) Â• " (C.4) Theorem: There exists two unitary transformations U and V having the property that if a' = aU (C.5) b' = bV then the overlap matrices S' = (a'b') (C.6) S't = (b'a') 1. See Appendix B. 100
PAGE 110
101 are going to be diagonal, leading to a natural "pairing" of the basis functions. Consider the two product matrices SS^ and S^S, which are Hermitian and of order mxm and n X n respectively. (If m > n then SS^ has at least m n eigenvalues equal to zero.^) Then there exists a unitary matrix V such that Vi(StS)V = u (C.7) where u is diagonal, and for the sake of simplicity, assumed to be of order n x n. (i.e. No vanishing eigenvalues.) Also uÂ„ > 0 since S+S is positive definite. As in relation (B.3) of Appendix B we define U = SVu~2 (C.8) where U^U = uW^S+SVu^ = u"^ uui = (C.9) and U+SV = uW+S^SV = u^ u = (C.IO) If for convenience we define Tr=4 (C.ll) then (a'rlb',) = TrSrs (C.12) 2. See Appendix B.
PAGE 111
102 as well as (bMa',) = TrSrs (C.13) which is the pairing theorem. The value of the overlap % ranges between 0 < Tr < 1.^ If = 1 then a'^ = b'^ and corresponds to a doubly occupied orbital. Those orbital pairs having Tr < 1 are referred to as the "corresponding orbitals" of Amos and Hall or the "paired orbitals" of Lowdin. Following the same arguments as those of Appendix B leads to the m n extra eigencolurans of SS^ associated with the m n zero eigenvalues, all necessarily being of zero length. 3. As long as the positive square roots of the Hk are chosen in relation C.8.
PAGE 112
APPENDIX D THE EVALUATION OF \^pns{0)f In Chapter 1 we introduced the topic we wished to investigate and the quantity we intended to compute, that being p^{r) of equation (1.10). In Chapters 2 and 3 we investigated methods for obtaining a "good" PÂ®;^ i.e. good in the sense of reproducing experimental observations. Here we shall be concerned with the evaluation of the probability, xjl(r) x^lr). of equation (1.89). Semiemperical methods in general, and the INDO/1 and INDO/S methods we investigate in particular, never explicidy define the atomic basis {x} [16]; rather they define Pock and other energy integrals. This leads to an interesting quandary; upon obtaining P'^, or P^, how does one go about obtaining (O) for the oneelectron operator O, (where O operates on the primed basis function, and 1' is set to 1 prior to integrating) when one does not have a basis set? Obviously without a {x} equation (D.l) is difficult to evaluate. It is customary in semiemperical theory to assume a basis set when seeking to evaluate (D.l), and that is what we do in this work. We select as our basis, a set of 1. See equation (2.63) and the following text. (D.l) 103
PAGE 113
104 symmetrically orthogonalized [54] Slater type atomic orbitals (STO) [29], XÂ„o = (Xl,X2. Xm) " (X..Â„X,,Â„) = S (D.2) or LOwdin type orbitals (LTO). We do so for the following two reasons, 1) to remain faithful to the INDO parameterization [16] and 2) Slater type orbitals behave as hydrogenlike atomic orbitals about the origin [14], i.e. a Slater type Is orbital is identical to a hydrogenlike Is orbital. All hydrogenlike stype orbitals have a proper cusp at the atomic origin. A plot of the nitrogen Is orbital can be found in Figure (7), where for Figures (7), (8) and (9), r(x, 0, 0) is defined in the following way, r(x,y,z) = i/x2 + y2 + z2 (D.3) r(x,0,0) = V^= x The unit of r is the bohr (oo) which is equal to 0.52918 x 10~^Â° m, the Is orbital exponent used in Figure (7) is taken from the work of Clementi [127], and has the value of 6.6651. The STO radial functions are nonorthogonal, their angular parts are the spherical harmonics [14], thus for orbitals on the same atomic center with differing angular momentum quantum numbers /, orthogonality is assured. Though those with the same / and m quantum numbers do have overlap. The normalized set of Slater orbitals are
PAGE 114
105 given as x:*;,.(r) = (2n)! 2n+l r"i exp(^r/ao) Y,(D.4) where n, / and m have their usual meaning, (i.e. The principal quantum number, n, the orbital angular momentum quantum number, /, and by convention, the z component of the orbital angular momentum, m.) The normalized spherical harmonics are given as Yr{0, 4>) = "2/ + 1 {l\m\)\ 47r (/+m)! P,'"*' (cos^) exp(Â— im^) (D.5) where (D.6) 2' /! ' rfiy'+lmi and It; = cos 0. Only 5type atomic orbitals, I = m Â— 0, "contact" the nucleus and these have no angular dependence. From equation (D.4), as well as from Figure (7), we see that the l5 Slater atomic orbital does indeed contact the nucleus, with a value of r = 0 (D.7) where Oo, the Bohr radius, is defined to be 1 in atomic units. For second period elements the value of xl'," at the origin is X',';{r) = NÂ„ r exp (^r) =4> 0, r = 0 (D.8) where 2n+l iVÂ„.= (20 47r(2n)! (D.9)
PAGE 115
106 this is true for all Slater 3type orbitals with n > 1, see equation (D.4). By Schmidt orthogonalizing the 25 to the Is we solve two problems, 1) the two radial functions are now orthogonal and 2) the Slater 25 atomic orbital now contacts its own nucleus. The orthogonal 2s' is given as (1 ALJ^ where A,,.. = (X: J X2,) (D.ll) and we have dropped the superscript sto from the {x}. The radial functions of the hydrogenlike atomic orbitals have n 1 nodes. It is obvious from (D.4) that the STO radial functions have no nodes. The orthonormalization has introduced the single node of the hydrogenlike 25 atomic orbital into the X2>'. Figures (8) and (9) are plots of 2 2 W2shyd{r)\ and X2.'('') , note the nodes and the behavior at the origin, and from Reference 14 3 V'2,hvd(r) = r( Â— ) (2Â— )exp(Zr/2ao) (D.12) where Z is the central charge. In Figure (8) Z of equation (D.12) has a value of 5.0, the core charge of a Nitrogen atom, assuming perfect shielding by the two I5 electrons. The Xz.' valence exponent, 6Â„ is that of nitrogen and has the value of 1.95, this from the work of Zemer et al
PAGE 116
107 [128, 129], and the Is exponent is the same as that used in Figure (7), i.e. 6.6651. As expected both functions are very similar in their description of the probability density, especially in the vicinity of the origin. This is the region we need to well describe when calculating the Fermi contact. The above diatribe would be irrelevant if all we wanted was a constant to multiply Tde/^egN^NP^,^, by, but we want more than that. We chose to use (D.7) and (D.IO) for two reasons, 1) we want to experiment with allowing the valence sorbital contact to vary as a function of the atomic charge, and so need a function to vary; and 2) we want to check the internal consistency of our model by using the observed aiao, the {x} and the computed Pf^^^ to obtain the "observed" valence Slater 5orbital exponent ^o6The closeness of (ob to ^p, the exponent used in the parameterization of the INDO model, is a measure of the compatibility of the INDO generated spin density and the "assumed" atomic basis. Tables (D1) and (D2) summarize the relationship between the used in the parameterization and those obtained when starting from the experimentally observed a.^o. i.e. The ^o* were obtained by subtracting the computed twoand threecenter contributions to a^Â„,, from the observed a.^o, and then solving equation (1.10) for ^06The inclusion of twoand threecenter terms in the computed a.so has the effect of slightly improving its calculated value for the tt radicals, and they must be included for the a radicals. The twoand threecenter contributions are large for a radicals, and the
PAGE 117
108 ^06 of the cr systems is far too large if only the onecenter terms are included.^ The ^( of both a and tt radicals are comparable if these contributions are included.^ Table D1: The of the standard Nitrogen optimization data set, where ^, = 1.95. System a^exp (G) Charge (au) T nwHin Charge (au) 1,4,5,8tetraazaanthracene Â±2.41 1.935 0.434 0.348 N,Ndihydropyrazine + + 7.60 1.972 0.222 0.020 pyrazine Â±7.22 1.985 0.532 0.467 N,Ndihydrophenazine + Â± 6.20 1.919 0.254 0.071 phenazine Â±5.14 1.875 0.500 0.432 N,Ndihydroquinoxaline + Â±6.65 1.921 0.236 0.048 quinoxaline Â±5.64 1.901 0.514 0.447 The Optimum Slater Valence Exponent We now seek the optimum valence 5orbital exponent for nitrogen and hydrogen. We label the optimized i as and they were determined in the following manor. A standard set of molecular systems was chosen (All experimental information can be found in the references listed in Table (D10).) and their structures optimized by either an INDO/1SCF method or an abinitio SCF method using a 63 IG** basis set.'* Then the aH and aN were calculated using equations (D.14) and (D.15) respectively. The were then 2. See Table (D9). 3. See the CIS values of Tables (D6). 4. See Table (D10).
PAGE 118
109 C/3 lH u I 00 b Q c 3 3 3 u c o a O Oh a a. H B S C/5 ^ On (S CM oo oo oo e c o o o d >n m rnooo\OinTHco 2 ^ 2 ^ On rt' of ^ m vo oo oo vo cn >n Â«n o CO CN NO IT) On ^ cn in c4 ^ +1 +1 +1 +1 +1 +1 +1 +1 .3 is 1^ O O m OO fO 00 On tÂ— I oo m
PAGE 119
110 determined such that the error between the calculated ciiso, Ocoio ^nd the observed ajso is at a minimum. The function minimized was the rootmeansquare of the normalized errors, i.e. /Â«) = y (D.13) N and the method used in finding the minimum was that of PolakRibiere [133, 134], a gradient based method. The functions used for aca/c(0 equation (D.13) are, for the valence Is and the 2s orbitals, respectively, " VÂ«..+6.)'i92ff.a7 where C;^ is a constant of proportionality specific to each nucleus, : C^ = 7jv(0.059621054) (D.16) and is the magnetogyric ratio [1] of nucleus N. The numerical constant of equation (D.16) incorporates the constants of equation (1.10), as well as conversion factors to relate Jn, expressed in Gaussian units, with and ( expressed in atomic units, to yield a result in Gauss. The nitrogen of (D.15) is again that of Clementi [127]. Equations (D.14) and (D.15) were obtained by the evaluation of V'i.(0)P and V'2.'(0)^ respectively [135]. Tables (D3) through (D5) catalog the results of the minimizations. Before we review
PAGE 120
111 the results we need to comment on an interesting observation. Figure (10) is a plot of vs Aiso, where A,so is defined here to be a^^p minus the calculated twoand threecenter contributions. It is obvious from Figure (10) that the cr and tt hydrogens are treated somewhat differently by the INDO Hamiltonian. We should note here that analogues plots for a abinitio UHF ST03G and 32 IG** basis, yield similar results. Part of the difference is due to the twoand threecenter contributions. The curve labeled 3CNT is a plot of a^^p vs /9f, where the twoand threecenter terms are accounted for; the curve labeled a does not have these terms taken into account, and the curve labeled tt is a plot of the TT radical a^^^ vs AisoFrom Table (D6) we see that the CIS hydrogen a and tt are quite similar, and while their plotted curves still differ, the (op for the a compounds has changed from 1.454 to 1.250 upon accounting for the twoand threecenter terms. Sigma systems interact directly, while for tt systems exchange is important. Multicenter exchange is missing in the INDO Hamiltonian, and this can be expected to have an effect on the c7 and TT slopes. Figure (14) clearly shows that once die twoand threecenter terms are accounted for, and is used, the apparent difference between the cr and tt systems is no longer there. We will address this further in the conclusions. The results of Tables (D6) and (D7) are obvious. For hydrogen and nitrogen the INDO/S Hamiltonian is superior to the INDO/1 Hamiltonian. Within the INDO/S model, ROHFCIS gives the best results. For these reasons we will present the INDO/S results in Chapters (2) and
PAGE 121
112 (3), and refrain from presenting the INDO/1 results. It is of interest to note that the hydrogen of Table (D6) approach the same value for (7 and tt radicals for the CIS and PUHF wavefunctions, but do not for the UHF wavefunction. It would seem the UHF wavefunction proportionately over estimates the spin density on the hydrogens lying in the nodal plane of tt radicals, while PUHF and CIS seem to get it proportionately right.
PAGE 122
113 Table D3: Minimization results for tt hydrogen. Optimized Valence 5orbital Exponent UHF PUHF CIS INDO/S (Error %) 40.2 40.2 6.14 Max Err % 98.7 98.0 11.2 INDO/1 (Error %) 42.8 42.8 18.0 Max Err % 96.5 96.4 32.5 Table D4: Minimization results for crhydrogen. Optimized Valence 5orbital Exponent UHF PUHF CIS INDO/S (Error %) 6.89 7.55 6.49 Max Err % 13.0 13.1 12.4 INDO/1 (Error %) 26.7 14.2 18.0 Max Err % 85.8 33.0 32.5 Table D5: Minimization results for tt nitrogen. Optimized Valence 5orbital Exponent UHF PUHF CIS INDO/S (Error %) 8.92 8.91 4.07 Max Err % 15.3 15.1 9.74 INDO/1 (Error %) 17.9 8.91 6.86 Max Err % 35.9 15.1 12.8
PAGE 123
114 Table D6: The optimized hydrogen exponents. Optimized Valence 5orbital Exponent UHF PUHF CIS a onecenter top 1.406 1.454 1.424 a one, twoand threecenter 1.252 1.295 1.274 a and tt one, two, and threecenter 0.961 1.389 1.248 Table D7: The optimized nitrogen exponents. Optimized Valence 5orbital Exponent UHF PUHF CIS 1.792 2.503 1.956 The Charge Dependence of i We had initially hoped to find a relationship between the computed charge of the atomic center, either the Mulliken charge [130, 131] or the Lowdin charge [132], and the "necessary" V'(0)p; i.e. the necessary to obtain aÂ„p. Such a relationship would have allowed the valence 5orbital to expand or contract, "breath", as a function of the charge associated with each magnetic center. Inspection of Tables (D1) and (D2) clearly shows that there is no such relationship, and we abandon this line of investigation here.
PAGE 124
115 Table D8: Computed UHF, PUHF and ROHFCIS results for the optimum Slater valence exponent Magnetic Molecular System Center' T TT TT" UHF TtT TT TT" PUHF CIS phenazine 0.847 H 3.66 1.70 1.15 Â± 1.93 0.45 0.21 0.70 Â± 1.61 N 6.27 6.11 5.10 Â± 5.14 N,Ndihydrophenazine + 0.783 H 5.38 5.44 7.38 Â± 6.79 1.12 1.19 1.28 + 1.77 0.44 0.30 0.23 Â± 0.59 N 6.36 6.32 6.10 + 6.20 quinoxaline 0.822 H 1.85 2.03 3.24 + 3.32 1.39 1.46 1.75 + 2.32 0.16 0.15 0.74 Â± 1.00 N 6.26 6.08 5.08 Â± 5.64 NJ>Jdihyclroqumoxalme + 0.767 H 5.90 5.94 8.14 + 7.17 2.56 2.61 4.67 Â± 3.99 0.88 0.92 1.01 Â± 1.38 0.27 0.34 0.26 Â±0.78 N 6.41 5.49 6.52 Â±6.65 oxacyclobutane + 0.751 H 66.5 75.5 63.7 63.0 12.2 14.2 11.6 11.0 tetrahydrofuran + 0.751 H 86.0 98.3 82.0 88.5 44.2 50.7 42.1 39.8 7. The isotopes are "iV and ^H.
PAGE 125
116 Table D9: The effects of including the twoand threecenter contributions to a Number of Included Terms Magnetic 1 Term 3 Terms System Center flcaic (G) acalc (G) Oexp (G) NO2 N 56.2 55.1 52.5 PH2 H 10.8 13.5 17.9 CeHg H 4.95 4.89 3.75 NH3 + N 20.2 21.1 19.5 H 21.9 25.5 25.9 C4H8O H 55.7 82.0 88.5 H 28.6 42.1 39.8 CsHeO H 44.7 63.2 63.0 H 8.37 11.61 11.0
PAGE 126
117 Table D10: The Optimization Data Set. Chemical Name Formula Charge Reference Optimized oxacyclobutane CsHeO +1 136 631G** tetrahydrofuran C4H8O +1 137 631G** anthracene C14H10 1 138 INDO/1 naphthalene CioHg 1 139 INDO/1 anthracene C14H10 +1 138 INDO/1 phenazine C12H8N2 1 140 INDO/1 quinoxaline CgHeNa 1 141 INDO/1 N,Ndihydropyrazine C4H6N2 +1 140 INDO/1 1 ,4,5,8tetraazaanthracene C10H6N4 1 141 INDO/1 pyrazine C4H4N2 1 141 INDO/1 nitrogen dioxide NO2 0 142 631G** nitric oxide NO 0 142 631G** benzene CeHg 1 142 INDO/1 ammonia NH3 +1 142 63 IG** N,Ndihydrophenazine C12H10N2 +1 143 INDO/1 N,Ndihydroquinoxaline C8H8N2 +1 143 INDO/1 The Results of Optimizing the Valence Exponent In Table (D8) we summarize the results of our optimization of the valence 5orbital exponents. It is clear that the smaller the hfc, the greater the absolute error. Overall though, the results are quite acceptable. Of course the real test will come when we apply the results of this appendix to the Bacteriochlorophyll and Bacteriopheophytin radical ions. The numbering for these structures can be found in Figure (15). From Table (D8) we see that while the magnitude of the splittings are "reasonable"
PAGE 127
118 for all these methods, the signs are suspect All 7rradical hydrogens are expected to have negative splittings, and as we can see only the ROHFCIS method predicts this for all such hydrogens. As we shall see later the ROHFCIS method always correctly predicts the sign right
PAGE 128
Figure 8: The Hydrogenlike 2s probability density.
PAGE 129
120 ' IJO IXK) 0^0 0.00 0.50 1.00 1.50 r(x,0,0) (bohr) Figure 9: The Slater type 2s' probability density. UMSJ ' II I 'III 0.01 0.00 0.01 0.02 0.03 0.04 0.05 0.06 One Center (p ) SS Figure 10: A plot of the spectroscopic QS P' vs the experimental A.v
PAGE 130
121 1.00 7.00 0.008 0.007 0.006 0.005 0.004 0.003 0.002 0.001 0.000 0.001 0.002 One Center (p^^ J Rgurc 11: The x region of Figure (10) expanded. 100.00 O 80 00 o Ji'' 60.00 20.00 INDO/SUHF Slope =1.01 IntCTcept = 0.99 20.00 0.00 20.00 40.00 60.00 80.00 Calculated A. (G) 100.00 ISO Figure 12: A plot of UHF aÂ„Â„ vs aÂ„,. using ^..= 0.961.
PAGE 131
122 20.00 0.00 20.00 40.00 60.00 80.00 100.00 Calculated A^^ (G) Figure 13: A plot of PUHF aÂ„,<, vs aÂ„Â„ using = 1.389. 20.00 I I 1Â— I.I.I Â—I . 1Â—^ 1^^ I . I 10.00 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00 Calculated A. ^ (G) Figure 14: A plot of aÂ„,, vs aÂ„,, using ^,,=1.248.
PAGE 132
Figure 15: The number system for Bacteiiochlon^hyll a The number system for Bacteriochlorophyll 6
PAGE 133
REFERENCES 1. A. Carrington and A. D. McLachlan, Introduction to Magnetic Resonance, Harper and Row, New York, 1967. 2. C. P. Slichter, Principles of Magnetic Resonance, Harper and Row, New York, 1963. 3. R. W. Fessenden and R. H. Schuler, in Advances in Radiation Chemistry, Vol. II, edited by M. Burton and J. L. Magee, page 1, Interstate, New York, 1970. 4. P. A. M. Dirac, Proc. Roy. Soc. (London) A117, 610 (1928). 5. P. A. M. Dirac, The Principles of Quantum Mechanics, Oxford University Press, London and New York, fourth edition, 1958. 6. J. Wertz and J. Bolton, Electronic Spin Resonance, McGrawHill, New York, 1972. 7. J. E. Harriman, Theoretical Foundations of Electron Spin Resonance, Academic Press, New York, 1st edition, 1978. 8. E. Fermi, Z. Physik. 60, 320 (1930). . . 9. W. Cheston, Elementary Theory of Electric and Magnetic Fields, John Wiley and Sons, New York, 1st edition, 1964. 10. M. Mizushima, Quantum Mechanics of Atomic Spectra and Atomic Structure, W. A. Benjamin Inc., New York, 1 edition, 1970. 11. A. Szabo and N. S. Ostlund, Modern Quantum Chemistry, Macmillian Publishing Co., London, 1st edition, 1982. 12. A. J. Stone, Proc. Roy. Soc. A271, 424 (1963). 13. H. A. Bethe and E. E. Salpeter, Quantum Mechanics of Oneand TwoElectron Atoms, Springer, Berlin, 1957. 124
PAGE 134
125 14. 1. Levine, Quantum Chemistry, AUyn and Bacon, Boston, 1974. 15. J. Lowe, Quantum Chemistry, Academic Press, New York, 1978. 16. J. A. Pople and D. L. Beveridge, Approximate Molecular Orbital Theory, McGrawHiU, New York, 1970. 17. in The Chlorophylls, edited by H. Scheer, CRC Press, Boca Raton, 1991. 18. in The Porphyrins Volume /V, edited by D. Dolphin, Academic Press, New York, 1979. 19. D. Hartree, Proc. Cambridge PhU. Soc. 24, 89 (1928). 20. J. Slater, Phys. Rev. 34, 1293 (1929). 21. V. Fock, Z. Physik 61, 126 (1930). 22. J. Slater, Phys. Rev. 35, 210 (1930). 23. C. C. J. Roothaan, Rev. Mod. Phys. 23, 69 (1951). 24. G. G. HaU, Proc. R. Soc. London Ser. A 205, 541 (1951). 25. C. C. J. Roothaan, Rev. Mod. Phys. 32, 179 (1960). 26. M. Bom and J. Oppenheimer, Ann. Physik 84, 457 (1927). 27. D. R. Hartiee, Proc. Camb. PhU. Soc. 89, 111 (1928). 28. F. E. Harris, UQC 13, 189 (1978). 29. J. Slater, Quantum Theory of Atomic Structure, McGrawHill, New York, 1960. 30. J. C. Slater, Phys. Rev. 34, 1293 (1929). 31. G. Uhlenbeck and S. Goudsmit, Naturwiss 13, 953 (1925). 32. G. Uhlenbeck and S. Goudsmit, Nature 117, 264 (1926).
PAGE 135
126 33. O. Stem, Z. Physik 7, 249 (1921). 34. W. Gerlach and O. Stem, Z. Physik 8, 110 (1922). 35. J. Pople, D. Santry, and G. Segal, J. Chem. Phys. 43, 5129 (1965). 36. J. Pople, D. Beveridge, and P. Dobosh, J. Chem. Phys. 47, 2026 (1967). 37. J. Sadlej, SemiEmperical Methods of Quantum Chemistry, John Wiley and Sons, New York, 1979. 38. J. J. P. Stewart, Seraiempireical molecular orbital methods, vol. 1, in Reviews in Computational Chemistry, edited by K. B. Lipkowitz and D. B. Boyd, pages 4582, VCH PubUshers Inc., New York, 1990. 39. M. C. Zemer, Semiempireical molecular orbital methods, in Reviews in Computational Chemistry, Vol. 2, edited by K. B. Lipkowitz and D. B. Boyd, pages 313366, VCH PubUshers Inc., New York, 1990. 40. J. A. Pople and R. K. Nesbet, J. Chem. Phys. 22, 571 (1954). 41. P. O. Lowdin, Phys. Rev. 97, 1509 (1955). 42. P. O. Lowdin, Advan. Phys. 5, 161 (1956). 43. A. T. Amos and G. G. Hall, Proc. Roy. Soc. (London) A263, 483 (1961). 44. J. G. W. Pratt, Phys. Rev. 102, 1303 (1956). 45. B. Sutcliffe, J. Chem. Phys. 39, 3322 (1963). 46. E. A. Hylleraas, Z. Phys. 54, 347 (1929). 47. C. Eckart, Phys. Rev. 36, 878 (1930). 48. T. Itoh and H. Yoshizumi, J. Phys. Soc. Japan 10, 201 (1955). 49. T. Itoh and H. Yoshizumi, J. Chem. Phys. 23, 412 (1955).
PAGE 136
127 50. H. H. Dearman and R. Lefebvre, J. Chem. Phys. 34, 72 (1961). 51. P. O. LQwdin, Adv. Chem. Phys. 2, 207 (1959). 52. K. F. Berggien and R. F. Wood, Phys. Rev. 130, 198 (1963). 53. P. O. Lowdin, J. Appl. Phys. 33, 251 (1962). 54. A. Szabo and N. S. Ostiund, Modern Quantum Chemistry. Introduction to Advanced Electronic Structure Theory, McGrawHill, New York, 1989. 55. P. O. Lowdin, Phys. Rev. 101, 1730 (1956). 56. T. Yonezawa, H. Nakatsuji, T. Kawamura, and H. Kato, J. Chem. Phys. 51, 669 (1969). 57. H. Nakatsuji, H. Kato, and T. Yonezawa, J. Chem. Phys. 51, 3175 (1969). 58. F. Sasaki and K. Ohno, J. Math. Phys. 7, 1140 (1963). 59. V. H. S. Jr., J. Chem. Phys. 41, 277 (1964). 60. P. O. Lowdin, Rev. Mod. Phys. 32, 328 (1960). 61. P. O. Lowdin, Quantum Theory of Atoms, Molecules and the Solid State, Academic Press, New York, 1966. 62. A. Hardisson and J. E. Harriman, J. Chem. Phys. 46, 3639 (1967). 63. P. O. Lowdin, Adv. in Phys. 5, 159 (1956). 64. R. Paunz, J. Chem. Phys. 37, 2739 (1962). 65. J. G. W. Pratt, Phys. Rev. 92, 278 (1953). 66. J. E. Harriman, J. Chem. Phys. 40, 2827 (1964). 67. J. H. Wilkinson and C. Reinsch, Linear Algebra Handbook for Automatic Compution, SpringerVerlag, New York, 1 edition, 1971.
PAGE 137
128 68. J. E. Haniman, Technical report. Quantum Chemistry Group, University of Uppsala, Sweden, 1963, Technical Note 97. 69. D. Phillips and J. Schug, J. Chem. Phys. 61, 1031 (1974). 70. K. M. Sando and J. E. Harriman, J. Chem. Phys. 47, 180 (1967). 71. R. Lefebvre, H. H. Dearman, and H. M. McConneU, J. Chem. Phys. 32, 1 (1960). 72. R. McWeeny and Y. Mizuno, Proc. Roy. Soc. A259, 554 (1961). 73. R. McWeeny, Rev. Mod. Phys. 32, 335 (1960). ' ' 74. J. Baker, J. Chem. Phys. 91, 1789 (1989). 75. J. Baker, Chem. Phys. Lett. 152, 227 (1988). 76. I. Mayer, Advan. Quant. Chem 12, 189 (1980). 77. F. Lendzian et al., Chem. Phys. Lett. 126, 290 (1986). 78. T. Amos and L. C. Snyder, J. Chem. Phys. 41, 1773 (1964). 79. L. C. Snyder and T. Amos, J. Chem. Phys. 42, 3670 (1965). 80. W. Edwards and M. Zemer, Theor. Chim. Acta 72, 347 (1987). 81. C. C. J. Roothaan, Rev. Mod. Phys. 32, 179 (1960). 82. E. Davidson, Chem. Phys. Lett. 21, 565 (1973). 83. C. Jackels and E. Davidson, Int. J. Quant. Chem. 8, 707 (1974). 84. A. Veillard, Computational Techniques in Quantum Chemistry, C. D. Reidel, Boston, 1975. 85. R. CabaUol, R. Gallifa, J. Riera, and R. Carbo, InL J. Quant. Chem. 8, 373 (1974). 86. R. Carbo, L. Domingo, and J. Gregori, Int. J. Quant. Chem. 17, 725 (1980).
PAGE 138
129 87. R. Carbo and J. Riera, A General SCF Theory, Lect. Notes in Ghent, Vol. 5, Springer, New York, 1978. 88. R. McWeeny and R. Dickerson, J. Chem. Phys. 49, 4852 (1968). 89. M. Guest and V. Sanders, Mol. Phys. 28, 819 (1968). 90. J. Binkley, J. Pople, and P. Dobosh, Mol. Phys. 28, 1423 (1974). 91. K. Faegri and R. Manne, Mol. Phys. 31, 1037 (1976). 92. H. Hsu, E. Davidson, and R. Pitzer, J. Chem. Phys. 65, 609 (1976). 93. R. Pauncz, Spin Eigenfunctions, Plenum Press, New York, 1979. 94. W. Ritz, J. Reine Angew. Math. 135, 1 (1909). 95. E. HyUeraas and B. Undheim, Z. Phys. 65, 759 (1930). 96. J. MacDonald, Phys. Rev. 43, 830 (1933). 97. N. Mataga and K. Nishimoto, Z. Physik. Chem. 13, 140 (1957). 98. H. Weyl, The Theory of Groups and Quantum Mechanics, Dover, New York, 1931. 99. L. Brillouin, Les Cahmps SelfConsistent de Hartree et de Fock, Hermann and Cie, Paris, 1934. 100. W. D. Edwards, B. Weiner, and M. Zemer, J. Am. Chem. Soc. 108, 2196 (1986). 101. W. D. Edwards, B. Weiner, and M. Zemer, J. Phys. Chem. 192, 6188 (1988). 102. M. G. Cory and M. Zemer, Chem. Rev. 91, 813 (1991). 103. N. G. J. Richards and M. G. Cory, IJQC Symp. 19, 65 (1992). 104. B. Levy and G. Berthier, Int. J. Quant. Chem. 2, 307 (1968). 105. W. Heitler and G. Rumer, Z. Phys. 68, 12 (1931).
PAGE 139
130 106. G. Rumer, Gottinger Nachr. 3, 337 (1932). 107. G. Rumer, E. TeUer, and H. Weyl, Gottinger Nachr. 3, 449 (1932). 108. I. Cooper and R. Mcweeny, J. Chera. Phys. 45, 226 (1966). 109. I. Cooper and R. Mcweeny, J. Chem. Phys. 45, 3484 (1966). 110. R. Manne and M. Zemer, Int. J. Quant. Chem. 19, 165 (1985). 111. J. Slater, Phys. Rev. 38, 1109 (1931). 112. C. Reeves, Common. ACM 9, 276 (1966). 113. B. SutcUffe, J. Chem. Phys. 45, 235 (1966). ^ 114. R. Fauncz, Alternant Molecular Orbital Method, Saunders, Philadelphia, 1967. 115. M. Simmonetta, E. Gianinetti, and I. Vandoai, J. Chem. Phys. 48, 1579 (1978). 116. M. Zemer, Interim report 1989, Technical report, Eastman Kodak Co., 1989. 117. M. Bersohn and J. C. Baird, An Introduction to Electron Paramagnetic Resonance, W. A. Benjamin, Inc., New York, 1 edition, 1966. 118. D. M. Chipman, J. Chem. Phys. 78, 3112 (1983). 119. S. A. Perera, J. D. Watts, and R. J. Bartlett, J. Chem. Phys. 100, 1425 (1994). 120. B. Engels, Chem. Phys. Lett. 179, 398 (1991). 121. O. Tapis and O. Goscinski, Mol. Phys. 29, 1653 (1975). 122. M. Karslon, A. R. Katritzky, M. Sezafran, and M. C. Zemer, J. Org. Chera. 54, 6030 (1989). 123. M. Karslon, A. Katritzky, and M. C. Zemer, J. Org. Chem. 56, 134 (1991). 124. M. Karslon and M. C. Zemer, J. Phys. Chem. 96, 6949 (1992).
PAGE 140
131 125. J. Yu and M. C. Zemer, J. Chem. Phys. 100, 1353 (1994). 126. P. O. Lowdin, Set theory and linear algebra some mathematical tools to be used in quantum theory. Technical report. University of Florida, Quantum Theory Project, 1976, Technical Note 490. 127. E. Clementi, J. Chem. Phys. 38, 2686 (1963). 128. J. E. Ridley and M. C. Zemer, Theo. Chem. Acta. 32, 111 (1973). 129. M. C. Zemer, G. H. Loew, R. F. Kirchner, and U. T. MuellerWesterhoff, J. Amer. Chem. Soc. 102, 589 (1980). 130. R. S. Mulliken, J. Chem. Phys. 23, 1833 (1955). 131. R. S. Mulliken, J. Chem. Phys. 23, 2338 (1955). 132. P. O. Lowdin, J. Chem. Phys. 18, 365 (1950). 133. E. Polak, Computational Methods in Optimization, Academic Press, New York, 1971. 134. W. H. Press, B. P. Flarmery, S. A. Teutolsky, and W. T. Vetterling, Numerical Recipes: The Art of Scientific Computing., Cambridge University Press, New York, 1986. 135. J. C. Slater, Quantum Theory of Atomic Structure, Volume I, McGrawHill, New York, 1960. 136. M. C. R. Symons and B. W. Wren, J. Chem. Soc. Perkins Trans. H 1984, 511 (1984). 137. H. Kubodera, T. Shida, and K. Shimokoshi, J. Chem. Phys. 85, 2583 (1981). 138. J. R. Bolton and G. K. Fraenkel, J. Chem. Phys. 40, 3307 (1964). 139. N. M. Atherton and S. I. Weissman, J. Am. Chem. Soc. 83, 1330 (1961). 140. J. R. Bolton, A. Carrington, and J. dos SantosVeiga, Mol. Phys. 5, 465 (1962).
PAGE 141
132 141. A. Carrington and J. dos SantosVeiga, Mol. Phys. 5, 21 (1962). 142. W. Weltner, Magnetic Atoms and Molecules, Scientific and Academic Editions, New York, 1 edition, 1983. 143. B. L. Barton and G. K. Fraenkel, J. Chem. Phys. 41, 1455 (1964).
PAGE 142
BIOGRAPHICAL SKETCH Marshall Cory was bom in Miami Florida, on July 4th, 1951 to Catherine and Marshall Cory, Sr. He has a sister Jeannine, some three and onehalf years his junior and a brother Sam, many years his junior. Marshall experienced the typical childhood of the era, being educated one part by television and one part by the Dade County school system. Once having graduated said school system he spent two years in the United States Marine Corps. Upon graduation from the Corps he decided to attend college and become the first member of his immediate family to do so. While in college he discovered he enjoyed both chemistry and physics, and graduated from Florida International University in June 1978, with a B.S. in chemistry and a minor in physics. In 1984 he decided to obtain a postgraduate degree in physical chemistry. Somewhere along the way to graduate school he acquired a wife, Genny, and an instant family, Lisa and Julie. He has enjoyed the time he spent at the University of Florida, which included both teaching and learning chemistry and physics. He feels it is now time to get on with the rest of his life, and is curiously looking forward to seeing how it turns out. 133
PAGE 143
I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Michael C. Zemepf^tairman Professor of Chemistry I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. N. Yngve 01 Professor of emistry and Physics I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. William Welmer, Jr. Professor of Chemistry I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fiilly adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. David Micha Professor of Chemistry and Physics
PAGE 144
I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Murali Rao Professor of Mathematics This dissertation was submitted to the Graduate Faculty of the Department of Chemistry in the College of Liberal Arts and Sciences and to the Graduate School and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. April 1994 Dean, Graduate School
xml version 1.0 encoding UTF8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchemainstance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID ECPMJ3E0V_YEAH4Z INGEST_TIME 20150401T19:33:40Z PACKAGE AA00030049_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES

