PARAMETERIZATION OF THE AM1* SEMIEMPIRICAL MOLECULAR ORBITAL METHOD FOR THE FIRST-ROW TRANSITION METALS AND OTHER ELEMENTS
Der Naturwissenschaftlichen Fakultät der Friedrich-Alexander-Universität Erlangen-Nürnberg
zur Erlangung des Doktorgrades Dr. rer. nat.
vorgelegt von Hakan Kayi aus Bursa, Türkei
2009
Als Dissertation genehmigt durch die Naturwissenschaftliche Fakultät der Friedrich-Alexander-Universität Erlangen-Nürnberg
Tag der mündlichen Prüfung: 23.12.2009
Vorsitzender der Promotionskommission: Prof. Dr. Eberhard Bänsch
Erstberichterstatter: Prof. Dr. Timothy Clark
Zweitberichterstatter: Prof. Dr. Nicolai Burzlaff
ACKNOWLEDGMENTS
First of all I would like to express my deepest gratitude to my supervisor Prof. Dr. Timothy Clark for his endless support, patience, full confidence in me and for always encouraging me. Without his understanding and excellent guidance, I could never complete this project. I am also very thankful for his support on personal level, that has been very valuable during many difficult situations.
I would like to thank Prof. Dr. Rudi van Eldik and Prof. Dr. Bernd Meyer for accepting to be examiners at my defense exam, and also Prof. Dr. Nicolai Burzlaff for his expertise.
I am very thankful to Dr. Paul Winget and Anselm Horn for sharing their great experiences in programming and for their help in using of parameterization scripts, and Dr. Nico van Eikema Hommes for his support and advice for solving technical problems related to hardware, software and scripting. Additionally, my thanks go to my officemate Dr. Tatyana Shubina for her fruitful discussions. I am also very thankful for the help and very valuable discussions of Dr. Matthias Henneman. Many thanks go as well to my former and present colleagues for their friendliness and support in many different issues: Dr. Harald Lanig, Dr. Ralph Puchta, Dr. Gudrun Schürer, Dr. Kendall Byler, Dr. Florian Haberl, Dr. Olaf Othersen, Dr. Jr-Hung Lin, Dr. Frank Beierlein, Dr. Mateusz Wielopolski, Dr. Pawel Rodziewicz, Dr. Gül Özpinar, Dr. Adria Gil Mestres, Dr. Ute Seidel, Christian Kramer, Christof Jäger, Sebastian Schenker, Matthias “Döner” Wildauer, Angela Götz, Marcel Youmbi and Alexander Urban. I also would like to say thank you to my friends: Kurtulus Erdogan, Günay Kaptan and Can Metehan Turhan for all the unforgettable moments they shared with me in Erlangen, and those everywhere that stayed friends despite the separations of time and distance.
For the financial support I thank Deutsche Forschungsgemeinschaft (GK312 “Homogeneous and Heterogeneous Electron Transfer” and SFB583 “Redox-Active Metal Complexes”).
And finally, I would like to express my warmest and endless gratitude to my parents and my sister for their lifelong love, support and encouragement.
i
ii
ZUSAMMENFASSUNG
In dieser Doktorarbeit wird die Parametrisierung der semiempirischen AM1* Molekülorbitaltechnik für einige neue Elemente beschrieben. Dies beinhaltet Resultate und Parameter für Vanadium, Chrom, Mangan, Eisen, Kobalt, Nickel, Kupfer, Zink, Brom, Jod und Gold. Die AM1* Methode ist eine Erweiterung der ursprünglichen AM1 Molekülorbital Theorie. Sie benutzt d-Orbitale für Elemente ab der zweiten langen Reihe des Periodensystems und eine leichte Modifizierung von Voityuk und Rösch’s AM1(d) Parametern für Mo. Unsere ursprüngliche Motivation für die Parametrisierung von AM1* war, die Vorteile von AM1 (gute H-Brücken Energien, höhere Rotationsbarrieren für π-Systeme als MNDO oder PM3) für die Elemente H, C, N, O und F beizubehalten, während die Performanz für P-, S- und Cl- beinhaltende Substanzen verbessert werden sollte. Zusätzlich wollten wir endlich eine Parametrisierung für Übergangsmetalle auf Basis eine MNDO Methode publizieren. Im Laufe der Arbeit stellte sich heraus, dass es auch nötig ist, Brom und Jod zu parametrisieren, um die Bromide und Jodide der Übergangsmetalle in den Parametrisierungsdatensätzen adäquat zu nutzen. Während der Vorbereitung der Parametrisierungsdatensätze wurden experimentelle Daten aus mehreren Quellen gesammelt. Im Falle fehlender experimenteller Daten und um den Bereich des Parametrisierungsdatensatzes zu vergrößern wurden die Eigenschaften wichtiger prototypischer Strukturen auf hoher quantenmechanischer Niveau berechnet. Zusätzlich haben wir experimentelle Daten geringer Qualität mit den Ergebnissen von hochqualitativen quantenmechanischen Rechnungen verglichen und verbessert. In solchen Fällen wurde das B3LYP Hybrid-Funktional mit LANL2DZ Basis Satz und Polarisierungsfunktionen oder Coupled Cluster Rechnungen mit Einzel- und Doppelanregungen und Störungstheoriekorrekturen für Dreifachanregungen (CCSD(T)) mit dem 6-311+G(d) Basissatz benutzt. Nach der Zusammenstellung des Datensatzes wurde die Parametrisierung durchgeführt und die Performanz und typische Fehler von AM1* analysiert. Diese werden im Detail gezeigt und mit der Performanz von anderen Methoden, die die NDDO Nährung benutzen, verglichen. Zusammenfassend lässt sich sagen, dass der AM1* Hamiltonian für energetische, elektronische und strukturelle Eigenschaften den anderen verfügbaren Hamiltonians überlegen ist, insbesondere für die Beschreibung von Übergangsmetallen.
iii
iv
ABSTRACT
This thesis describes the parameterization of AM1* semiempirical molecular orbital technique for a series of new elements. Parameterization results for vanadium, chromium, manganese, iron, cobalt, nickel, copper, zinc, bromine, iodine and gold are reported. The AM1* methodology is an extension of the original AM1 molecular orbital theory uses d-orbitals for the elements starting from the second long row of the periodic table, and a slight modification of Voityuk and Rösch’s AM1(d) parameters for Mo. Our original motivation in parameterizing AM1* was to retain the advantages of AM1 (good energies for hydrogen bonds, higher rotation barriers for π-systems than MNDO or PM3) for the elements H, C, N, O and F and to improve performance over AM1 for P-, S- and Cl-containing compounds and eventually to produce a published parameterization for an MNDO-like method for the transition metals. Additionally, bromine and iodine have also been parameterized because parameters for these elements became necessary in order to be able to parameterize the transition metals adequately by including their bromides and iodides in the parameterization datasets. In the preparation of parameterization datasets, experimental target data were collected from several sources. In the case of lack of experimental data and also to extend the range of parameterization dataset, prototype compounds were used and their properties derived from high-level calculations. In addition to this, when we have determined that the available experimental data are of insufficient quality, we have also applied corrections to available values using results from high-level calculations. In such cases, the B3LYP hybrid functional with the LANL2DZ basis set including polarization functions or coupled cluster calculations with single and double excitations and a perturbational corrections for triples (CCSD(T)) with the 6-311+G(d) basis set were used and dataset became more reliable. Once the parameterization dataset has been assembled, the parameterization process is performed. After this parameterization process is successfully completed, the performance and the typical errors of AM1* are discussed and also compared with the other available neglect of diatomic differential overlap (NDDO) Hamiltonians. To summarize, the performance of the AM1* for energetic, electronic and structural properties is superior to other available Hamiltonians in many cases, especially for the transition metals.
v
vi
CONTENTS
1 INTRODUCTION 1 1.1 Historical Development of Semiempirical Methods ...... 1 1.1.1 Hückel (HMO), Pariser-Parr-Pople (PPP) and Extended Hückel Theory (EHT) ..... 1 1.1.2 CNDO, INDO ...... 1 1.1.3 NDDO ...... 2 1.1.4 MINDO ...... 3 1.1.5 MNDO ...... 3 1.1.6 AM1 ...... 4 1.1.7 PM3 ...... 5 1.1.8 MNDO/d ...... 6 1.1.9 SAM1 ...... 6 1.1.10 PM3(tm) ...... 7 1.1.11 AM1(d) ...... 7 1.1.12 PM5 ...... 7 1.1.13 AM1* ...... 8 1.1.14 RM1 ...... 8 1.1.15 PM6 ...... 8 1.1.16 OMx-D ...... 9 1.2 Transition Metals ...... 9
2 THEORY 13 2.1 General Approximations ...... 13 2.2 Methods and Hamiltonians ...... 17 2.2.1 MNDO ...... 17 2.2.2 AM1 ...... 21 2.2.3 PM3 ...... 23 2.2.4 PM6 ...... 25 2.2.5 AM1* ...... 27
vii
3 PARAMETERIZATION OF SEMIEMPIRICAL METHODS 31 3.1 Introduction to Parameterization Methodology ...... 31 3.2 Reference Data ...... 31 3.3 Relative Weights of the Properties ...... 33 3.4 Parameter Optimization Procedure ...... 34
4 RESULTS OF AM1* PARAMETERIZATIONS 37 4.1 Parameterization Data ...... 37 4.2 Parameterization of Vanadium and Chromium ...... 38 4.2.1 Results ...... 38 4.2.1.1 Vanadium ...... 40 4.2.1.1.1 Heats of Formation of Vanadium Compounds ...... 40 4.2.1.1.2 Ionization Potentials and Dipole Moments of Vanadium Compounds ..... 46 4.2.1.1.3 Geometries of Vanadium Compounds ...... 47 4.2.1.2 Chromium ...... 55 4.2.1.2.1 Heats of Formation of Chromium Compounds ...... 55 4.2.1.2.2 Ionization Potentials and Dipole Moments of Chromium Compounds .... 60 4.2.1.2.3 Geometries of Chromium Compounds ...... 61 4.2.2 Conclusions and Outlook ...... 67 4.3 Parameterization of Manganese and Iron ...... 68 4.3.1 Results ...... 68 4.3.1.1 Manganese ...... 69 4.3.1.1.1 Heats of Formation of Manganese Compounds ...... 69 4.3.1.1.2 Ionization Potentials and Dipole Moments of Manganese Compounds ... 73 4.3.1.1.3 Geometries of Manganese Compounds ...... 74 4.3.1.2 Iron ...... 78 4.3.1.2.1 Heats of Formation of Iron Compounds ...... 78 4.3.1.2.2 Ionization Potentials and Dipole Moments of Iron Compounds ...... 84 4.3.1.2.3 Geometries of Iron Compounds ...... 85 4.3.2 Conclusions and Outlook ...... 93 4.4 Parameterization of Cobalt and Nickel ...... 94 4.4.1 Results ...... 94 4.4.1.1 Cobalt ...... 95 4.4.1.1.1 Heats of Formation of Cobalt Compounds ...... 95
viii
4.4.1.1.2 Ionization Potentials and Dipole Moments of Cobalt Compounds ...... 100 4.4.1.1.3 Geometries of Cobalt Compounds ...... 101 4.4.1.2 Nickel ...... 108 4.4.1.2.1 Heats of Formation of Nickel Compounds ...... 108 4.4.1.2.2 Ionization Potentials and Dipole Moments of Nickel Compounds ...... 113 4.4.1.2.3 Geometries of Nickel Compounds ...... 114 4.4.2 Conclusions and Outlook ...... 123 4.5 Parameterization of Copper and Zinc ...... 124 4.5.1 Results ...... 124 4.5.1.1 Copper ...... 125 4.5.1.1.1 Heats of Formation of Copper Compounds ...... 125 4.5.1.1.2 Reaction Energies of Copper Compounds ...... 127 4.5.1.1.3 Ionization Potentials and Dipole Moments of Copper Compounds ...... 128 4.5.1.1.4 Geometries of Copper Compounds ...... 129 4.5.1.2 Zinc ...... 133 4.5.1.2.1 Heats of Formation of Zinc Compounds ...... 133 4.5.1.2.2 Ionization Potentials and Dipole Moments of Zinc Compounds ...... 137 4.5.1.2.3 Geometries of Zinc Compounds ...... 139 4.5.2 Conclusions and Outlook ...... 145 4.6 Parameterization of Bromine and Iodine ...... 146 4.6.1 Results ...... 146 4.6.1.1 Bromine ...... 147 4.6.1.1.1 Heats of Formation of Bromine Compounds ...... 147 4.6.1.1.2 Ionization Potentials and Dipole Moments of Bromine Compounds ..... 153 4.6.1.1.3 Geometries of Bromine Compounds ...... 156 4.6.1.2 Iodine ...... 160 4.6.1.2.1 Heats of Formation of Iodine Compounds ...... 160 4.6.1.2.2 Ionization Potentials and Dipole Moments of Iodine Compounds ...... 164 4.6.1.2.3 Geometries of Iodine Compounds ...... 167 4.6.2 Conclusions and Outlook ...... 170 4.7 Parameterization of Gold ...... 171 4.7.1 Results ...... 171 4.7.1.1 Heats of Formation of Gold Compounds ...... 172 4.7.1.2 Ionization Potentials and Dipole Moments of Gold Compounds ...... 174
ix
4.7.1.3 Geometries of Gold Compounds ...... 176 4.7.2 Conclusions and Outlook ...... 181
BIBLIOGRAPHY 183
x
xi
xii
INTRODUCTION Chapter 1
1 INTRODUCTION
1.1 Historical Development of Semiempirical Methods
1.1.1 Hückel (HMO), Pariser-Parr-Pople (PPP) and Extended Hückel Theory (EHT)
The history of semiempirical methods started with theories based on purely π-electron treatments in the 1930s. The Hückel Molecular Orbital (HMO) method is the earliest, simplest and the most prominent π-electron theory for treating conjugated molecules [1]. It was used to predict the properties and reactivities of planar conjugated compounds. A great failing of HMO is its treatment of electron repulsion. The first semiempirical π-electron theory that includes formally the effect of electron repulsion between valence electrons and thence improves on HMO is the Pariser-Parr-Pople (PPP) method [2, 3, 4, 5, 6, 7]. Both HMO and PPP methods are only applied to planar conjugated molecules, but PPP allows heteroatoms other than hydrogen. PPP is popular for developing simple parameterized analytical expressions for molecular properties. Today it is still used in the cases that require minimal electronic effects. Extended Hückel Theory (EHT) appeared as a molecular orbital theory that takes into account all valence electrons in the molecule and is applicable to non-planar molecules [8, 9]. Since EHT is applicable to all periodic table elements, today there is still interest in EHT, especially for modeling inorganic compounds in a reasonable CPU time. Currently, EHT is also preferred for computing band structures, which are generally considered to be very computation-intensive calculations. One the other hand, one must consider that EHT is very poor at predicting molecular geometries.
1.1.2 CNDO, INDO
Later, all-valence-electron semiempirical methods using the Zero Differential Overlap (ZDO) approximation, such as Complete Neglect of Differential Overlap (CNDO) [10, 11, 12, 13], Intermediate Neglect of Differential Overlap (INDO) [10, 14] and Neglect of Diatomic Differential Overlap (NDDO) [10, 11] were also proposed by Pople and his collaborators starting from the mid-1960s. These self-consistent field models are related to the extended 1
INTRODUCTION Chapter 1
Hückel method in much the same manner as the Pople-Pariser-Parr method is related to the Hückel molecular orbital approximation [15]. Simply, in the ZDO approximation, all products of basis functions on the same electron coordinates are neglected when they are located on different atoms. CNDO is the simplest of the all-valence-electron NDO models. In this model, only valence electrons are explicitly treated, the inner-shell electrons are taken as a part of the atomic core [16]. The CNDO method has proven useful for some hydrocarbon results but little else. CNDO is still sometimes used to generate the initial guess for ab initio calculations on hydrocarbons [17]. In the INDO approximation, the primary modification to the CNDO approximation is that one-center repulsion integrals between atomic orbitals on the same atom are not neglected. However, the INDO approximation shares with CNDO the inadequate representation of electron repulsions involving atomic orbitals with directional properties. Today, the INDO method has largely been superseded by more accurate methods. It is still sometimes used as an initial guess for ab initio calculations. In 1973, the ZINDO program package, which contains INDO/1 to calculate molecular geometries and INDO/S especially designed to calculate electronic spectra of organic molecules and transition metal complexes, was introduced by Zerner’s group [18]. The program also predicts UV transitions well except metals with unpaired electrons. However, it produces generally poor results for geometry optimizations. An interesting intermediate neglect of differential overlap based technique, symmetrically orthogonalized INDO, in short SINDO and later SINDO1, is conceptually and practically superior to the INDO method [19, 20, 21]. The most important features of SINDO are that it explicitly takes ortogonalization transformations of the basis functions into account and treats inner orbitals by a local pseudopotential. SINDO appears to perform well but has not found the wide range of acceptance of the NDDO based methods.
1.1.3 NDDO
The NDDO method is an improvement on INDO, since it neglects differential overlap only when the atomic orbitals are on different atoms. Thus dipole-dipole interactions are retained and expressed in terms of integrals that are calculated either from atomic orbitals or determined empirically [15]. A few attempts were made to parameterize NDDO, but the results were rather disappointing [22].
2
INTRODUCTION Chapter 1
1.1.4 MINDO
A few years later, several the modified intermediate neglect of differential overlap methods (MINDO/1, MINDO/2, MINDO/3) were introduced by Dewar and coworkers [23, 24, 25]. Dewar aimed to calculate ground-state properties, in particular heats of formation and molecular geometries with chemical accuracy, such as bond lengths of 0.1 pm, bond angles of 0.1°, heats of formation that are correct to 0.1% and so on, taking all valence electrons into account [23]. His motivation for modifying INDO was to remove several deficiencies of INDO, especially in the analytically calculation of the one-electron repulsion integrals. He rather evaluated these integrals by using parameters and fitting these parameters to experimental data. Incorporating the Davidson-Fletcher-Powell geometry optimization routine [26], the parameterization program was able to accept initial geometries as input and derive the associated minimum energy structures. In this way, the MINDO methods became convenient for geometry optimization calculations. Here one can say that MINDO represented a very big step toward encouraging chemists to use molecular orbital calculations in the interpretation of experimental data. The third version of MINDO was by far the most reliable and was accepted to be the first modern semiempirical method. Though MINDO/3 has been superseeded by more accurate methods today, it is still sometimes used to calculate an initial guess for ab initio calculations. MINDO methods had some limitations, such as too positive heats of formation for unsaturated molecules, too large bond lengths and too negative heats of formation for molecules that contain adjacent atoms with lone pairs. Some of these limitations resulted from using the INDO approximation, particularly from the inability of INDO to deal with systems containing lone pairs [27].
1.1.5 MNDO
To overcome these limitations, in 1977 the modified neglect of diatomic overlap (MNDO) method was introduced by Dewar and Thiel [28]. This method is the oldest NDDO-based model that evaluates one-center two-electron integrals based on spectroscopic data for isolated atoms, and evaluates other two-electron integrals using the idea of multipole- multipole interactions from classical electrostatics. In MNDO, various integrals are not determined analytically, rather numerical parameters are adjusted to fit the experimental data as in MINDO. MNDO was mainly parameterized to reproduce heats of formation and the
3
INTRODUCTION Chapter 1
geometrical properties of stable molecules using ionization potentials and dipole moments as ancillary data. Possibly the most important advantage of MNDO over MINDO/3 is the use of monatomic parameters, while MINDO/3 requires diatomic parameters in resonance integrals and core-core repulsion. There are many articles available that compare the performance of MNDO with different semiempirical methods [29, 30, 31, 32, 33, 34, 35, 36]. MNDO has some known deficiencies, such as its inability to model intermolecular systems containing hydrogen bonds accurately when the atoms are separated by a distance around sum of their van der Waals radii. Additionally, hypervalent molecules are too unstable, four-membered rings are too stable, rotational barriers are often underestimated, activation barriers are too high, electronic excitation energies are underestimated, conformational preferences are sometimes not reproduced [37]. The major problem of MNDO, its tendency to overestimate repulsions between atoms at approximately their van der Waals distance was sought to be solved by modifying the core repulsion function in MNDO.
1.1.6 AM1
In this way, beside the MNDO/H, which was introduced by Burstein and Isaev to investigate H-bonded systems [38], Dewar and coworkers introduced their new method AM1 [39]. AM1 is basically a modification to and a reparameterization of the general theoretical model found in MNDO. Its major difference is the addition of Gaussian functions to the description of core repulsion function to overcome MNDO’s hydrogen bond problem. Additionally, since the computer resources were limited in 1970s, in MNDO parameterization methodology, the overlap terms, βs and βp, and Slater orbital exponents ζs and ζp for s- and p- atomic orbitals were fixed. That means they are not parameterized separately just considered as βs = βp, and
ζs = ζp in MNDO. Due to the greatly increasing computer resources in 1985 comparing to 1970s, these inflexible conditions were relaxed in AM1 and then likely better parameters were obtained. The addition of Gaussian functions significantly increased the numbers of parameters to be parameterized from 7 (in MNDO) to 13-19, but AM1 represents a very real improvement over MNDO, with no increase in the computing time needed. Dewar also concluded that the main gains of AM1 were its ability to reproduce hydrogen bonds and the promise of better estimation of activation energies for reactions [39]. However, AM1 has some limitations. Although hypervalent molecules are improved over MNDO, they still give larger errors than the other compounds, alkyl groups are too stable, nitro compounds are too unstable, peroxide bond are too short [40]. AM1 has been used very widely because of its 4
INTRODUCTION Chapter 1
performance and robustness compared to previous methods. This method has retained its popularity for modeling organic compounds and results from AM1 calculations continue to be reported in the chemical literature for many different applications.
1.1.7 PM3
In 1989, Stewart introduced PM3 [41, 42, 43], which can be considered as a reparameterization of AM1. This method was named as parametric method 3, considering MNDO and AM1 as the methods 1 and 2, respectively, as one of the three NDDO-based methods. In both MNDO and AM1, one-center electron repulsion integrals (g ij , h ij ), which are five parameters gss , gsp , gpp , gp2 , and gsp , are assigned values determined from atomic spectra by Oleari [44]. PM3 differs from MNDO and AM1 and these one-center electron integrals are taken as parameters to be optimized. PM3 also differs from AM1 in the number of Gaussian terms used in the core repulsion function. PM3 uses only two Gaussian terms per atom instead of up to four used by AM1. Another difference is that PM3 uses an automated parameterization procedure, in contrast to AM1. H, C, N, O, F, Al, Si, P, S, Cl, Br, and I parameters were simultaneously parameterized, whereas AM1 parameters were adjusted manually by Dewar with the help of chemical knowledge or intuition. Since his parameter optimization algorithms permitted an efficient search of parameter space, Stewart was able to employ a significantly larger data set in evaluating his penalty function than had been true for previous efforts [45]. Statistically, PM3 was more accurate than the other semiempirical methods available at the time [46], but it was found to have several deficiencies that seriously limited its usefulness. One of the most important of these is the rotational barrier of the amide bond, which is much too low and in some cases almost non-existent. The other one is that PM3 has a very strong tendency to make the environment around nitrogen pyramidal. Thus, PM3 is not suggested for use in studies where the state of hybridization of nitrogen is important [46].
According to a search of “Current Contents” done in 1999, AM1 was the most widely used semiempirical quantum mechanical method and PM3 was second [47].
5
INTRODUCTION Chapter 1
1.1.8 MNDO/d
The MNDO, AM1 and PM3 methods use only sp -basis sets and do not include d-orbitals in their original implementation. Therefore, these methods are unable to treat most of the transition metal compounds and a large part of the periodic table. Also, from ab initio calculations it is known that d-orbitals are significant for quantitative accuracy in the hypervalent compounds of main group elements. Because of these limitations and deficiencies, an extension of the MNDO formalism to d-orbitals was necessary to investigate transition metals and heavier main group elements. In 1992, Thiel and Voityuk introduced MNDO/d, the first NDDO model with d-orbitals [48, 49, 50]. MNDO/d explicitly contains d- orbitals for heavier atoms starting from the second long row in periodic table. It uses theory and parameters in original MNDO methodology unchanged for the elements where Z<11. In MNDO/d two-center two-electron integrals are calculated using the original point-charge model [51] which is also used in MNDO, AM1 and PM3. These integrals are expanded in terms of semiempirical multipole-multipole interactions. All monopoles, dipoles and quadrupoles of these charge distributions are included, whereas all higher multipoles beyond order 4 are neglected [52]. MNDO/d showed enormous improvements over MNDO, AM1 and PM3, especially for hypervalent molecules. MNDO and AM1 were not designed to treat hypervalent compounds, but in the parameterization of PM3 considerable effort was made to overcome such deficiencies. MNDO/d predicts the point groups of hypervalent compounds more accurately and also predicts the heats of formation of hypervalent compounds with small mean absolute errors compared to MNDO, AM1 and PM3 [34, 48, 49, 50]. Nonetheless one must consider that MNDO/d produces identical results to MNDO for the elements with Z<11.
1.1.9 SAM1
The last semiempirical technique produced by Dewar’s group is Semi-ab initio model 1 (SAM1) [53, 54, 55]. In this method, two-electron repulsion integrals are ab initio integrals that are evaluated from contracted Gaussian basis functions (STO-3G) fit to Slater-type orbitals using standard methods [56]. SAM1 is apparently quite successful for transition metals and vibrational frequencies, but neither the complete method nor a comprehensive analysis of its performance has been published. Due to the need of calculating two-electron repulsion integrals correctly in ab initio fashion, SAM1 is slower than AM1 but much faster
6
INTRODUCTION Chapter 1
than ab initio methods because it still uses the NDDO approximation. This method is only available within the commercial software package, AMPAC 9 [57].
1.1.10 PM3(tm)
The extension of MNDO formalism to d-orbitals not only formed a basis for MNDO/d but also for the independent PM3(tm) parameterization [58]. PM3(tm) has not been fully described in the literature, so it is one of those unpublished methods, and only available within commercial software package SPARTAN 8.0 [59]. Hehre and coworkers added d- orbitals to Stewart’s PM3 and parameterized only to reproduce the geometries of crystal structures of transition metal complexes. More usual properties, energies, dipole moments and ionization potentials were not taken into account for the parameterization. PM3(tm) appears not to be as reliable as hoped [60] but can be used to generate reasonable molecular geometries, whose energies may then be calculated with more reliable methods.
1.1.11 AM1(d)
In 2000, Voityuk and Rösch introduced the AM1(d) parameterization for molybdenum [61]. They extended AM1 to an spd -basis by adding d-orbitals to molybdenum. The core-repulsion function was also modified. They excluded Gaussian functions from the core-core repulsion term and included two new bond-specific parameters. The established AM1 formalism and all the parameters for all-main group elements were taken unchanged. Due to this fact, AM1(d) produces identical results to the original AM1 method for non-transition metal elements.
1.1.12 PM5
In addition to PM3(tm), another extension of PM3 molecular orbital technique, namely PM5, has also been produced by Stewart. PM5 is a reparameterization of PM3 to improve its performance. It has been parameterized for all main group elements and many of transition metals. This method is available within the commercial software LinMopac2002 [62]. The complete method has never been published. However, some articles that include the results from PM5 calculations and compare its performance are available [63, 64, 65, 66, 67, 68]. Today this method has been superseded by PM6.
7
INTRODUCTION Chapter 1
1.1.13 AM1*
In 2003, Clark and coworkers introduced AM1* [69], as an extension to AM1 molecular orbital method. AM1* is based on AM1, rather than MNDO or PM3, because AM1 reproduces the energies of hydrogen bonds relatively well and generally performs better for rotational barriers of partial double bonds than the other two methods. AM1* uses the parameters and theory of original AM1 method unchanged for the elements H, C, O, N and F, and adds d-orbitals to other elements using a modified core-repulsion function. The use of these original AM1 parameterization elements obviously limits AM1*’s ultimate accuracy in some cases. On the other hand, the recent parameterizations for AM1*, in particular transition metal parameterizations [66, 68, 70], have shown that AM1* performs very well compared to other available modern methods, PM5 and PM6.
1.1.14 RM1
In 2006, RM1 (Recife Model 1, which takes its name from the city Recife, Brazil) was introduced [71]. Without making any changes to original AM1 formalism and to the set of approximations used in original AM1 methodology, ten elements, H, C, N, O, F, P, S, Cl, Br, and I were reparameterized. This resulted with improved accuracy over all other NDDO methods for organic compounds.
1.1.15 PM6
Stewart introduced his most recent model, PM6, in 2007 [72]. He made several modifications to the core-core interaction term and also to the method of parameter optimization. Additionally, d-orbitals were used. PM6 has been parameterized for 70 elements and so provides very large application area. Modification of core-core interaction term has resulted in a significant improvement for main group elements and also the use of d-orbitals has allowed this method to be extended to the transition metals.
8
INTRODUCTION Chapter 1
1.1.16 OMx-D
Thiel and coworkers have continued to modify MNDO over the years, using effective core potentials for the inner orbitals and adding orthogonalization corrections as in SINDO1. They have introduced OM1, OM2 and OM3 models (OMx) [73, 74, 75, 76, 77, 78]. These corrections are important especially to describe torsional angles correctly. In 2008, they added dispersion correction terms taken from density functional work without modifying the standard OMx parameters and achieved significant improvements for non-covalent interactions in biochemical systems [79]. McNamara and Hillier [80] also incorporated dispersion terms into AM1 and PM3 just before OMx-D, taking the parameters for dispersion functions from the BLYP-D parameterization. And method performs well for the non- covalent complexes.
Other semiempirical methods exist than the models only presented here. Other methods and variations of these models or many of hybrid-models are also available today. Semiempirical methods keep developing steadily. The availability of different methods with a large variety of parameters provides a good starting point for future developments and reaction-specific local parameterizations and also comparison calculations.
1.2 Transition Metals
Transition metals are distinguished from the other elements by the presence of filled or partially filled d-orbitals in their valence shell in one or more of their oxidation states. For the first row transition metals, 3d -orbitals are filled, whereas 4d - and 5d -orbitals are filled for second row and third row transition elements, respectively. They are the elements of the groups 3-12 of the periodic table. However, especially group 12 and to some extent group 3 show significant analogy to main-group elements. Although scandium and zinc are in the d- block, they are not considered to be transition elements, generally. Because they have no ion that has a partially filled d-orbital. Electronic configurations of transition metals given below in Table 1.1.
9
INTRODUCTION Chapter 1
Table 1.1: The electronic configurations of transition metals 1st Row 2nd Row 3rd Row Scandium [Ar]3d l4s 2 Yttrium [Kr]4d l5s 2 Lanthanum [Xe]5d 16s 2
Titanium [Ar]3d 24s 2 Zirconium [Kr]4d 25s 2 Hafnium [Xe]4f 14 5d 26s 2
Vanadium [Ar]3d 34s 2 Niobium [Kr]4d 45s l Tantalum [Xe]4f 14 5d 36s 2
Chromium [Ar]3d 54s 1 Molybdenum [Kr]4d 55s 1 Tungsten [Xe]4f 14 5d 46s 2
Manganese [Ar]3d 54s 2 Technetium [Kr]4d 55s 2 Rhenium [Xe]4f l4 5d 56s 2
Iron [Ar]3d 64s 2 Ruthenium [Kr]4d 75s l Osmium [Xe]4f 14 5d 66s 2
Cobalt [Ar]3d 74s 2 Rhodium [Kr]4d 85s 1 Iridium [Xe]4f 14 5d 76s 2
Nickel [Ar]3d 84s 2 Palladium [Kr]4d l0 5s 0 Platinum [Xe]4f 14 5d 96s 1
Copper [Ar]3d 10 4s 1 Silver [Kr]4d 10 5s 1 Gold [Xe]4f l4 5d l0 6s l
Zinc [Ar]3d 10 4s 2 Cadmium [Kr]4d l0 5s 2 Mercury [Xe]4f 14 5d 10 6s 2
The chemistry of the transition metals has been examined for about two centuries and experimentalists and theoreticians have had a growing interest in the past three decades. Transition metals are very widely used in many important areas of chemistry. Especially coordination compounds (or complexes) of transition metals with organic reagents play a very important role in organometallic chemistry. These compounds behave differently to both ionic and covalent compounds in organic chemistry. Because of their complicated constitution, these compounds are considered to be complexes [81]. Compounds containing transition metals are also very important in biochemistry and they are generally responsible for the specific functionality of many enzymes. Transition metals are the active sides in many molecules relevant in catalytic processes, and also clusters of transition metals mediate catalytic surface processes in many reactions [82].
Calculations of transition metals are important to test the performance of the available quantum chemical theories. Many of the transition metals contain unpaired electrons in their valence shell. The determination of the correct spin states, which directly affects the properties, makes the calculation of transition metals difficult [83]. Generally the most common methods for calculating transition metal compounds are classical ab initio techniques and DFT. The use of effective core potentials for describing transition metals has become popular [84]. Although relativistic effects can be considered very conveniently and the number of electrons to be calculated is reduced, these methods are still very demanding and expensive for the calculation of transition metal compounds. Hartree-Fock-based ab initio
10
INTRODUCTION Chapter 1
methods are often used in the calculation of transition metals, but the HF approximation has been recognized as being in error for the first row transition metals with partly filled d-orbitals [85], while d 0 and d 10 transition metal compounds are reliable [86]. On the other hand, in using of DFT methods, the choice of the proper basis set, exchange and correlation functionals is a very critical point and directly affects the quality of the results.
There are also several semiempirical methods available for the treatment of transition metals efficiently. One of the oldest of those, INDO/S [18] was especially designed to calculate electronic spectra of the transition metal complexes beside organic compounds. However, it is not used in the prediction of energetic and geometrical properties since it is not reliable for these properties. SINDO1 [19, 20, 21] has been extended to some transition metals. This method is not able to treat open shell systems and its accuracy is within the limits of classical INDO [10, 14] approximation.
Today most of the available semiempirical methods to treat transition metals are the ones based on the NDDO [10, 11] approximation. An extension of MNDO [28] method to d- orbitals, MNDO/d [48, 49, 50], was the first method to include d-orbitals that were necessary to investigate transition metals. However, only zinc, cadmium and mercury parameters were published for MNDO/d method [34]. An unpublished method, Semi-ab initio model 1 (SAM1) with a different theoretical basis, is quite successful for transition metals and vibrational properties, but the only transition metals available for SAM1 are iron and copper [53, 54, 55]. As an extension of PM3 to d-orbitals, PM3(tm) was parameterized to reproduce the geometries of transition metal compounds [58]. PM3(tm) parameters are available for Ti- Zn, Zr, Mo, Ru-Pd, Cd, Hf, Ta, W, Hg and Gd. But this method is not able to reproduce energetic and electronic properties, which are very important. AM1(d), which is an extension of the AM1 method to d-orbitals, is also available for molybdenum [61]. Another one of the unpublished method, PM5 [62] was parameterized for many of the transition metals, but has been superseded by PM6 [72]. PM6 contains all d-block elements and therefore provides a very wide application area. And finally, AM1*, which is again as an extension of AM1 to d- orbitals, has been parameterized for all first row transition metals plus gold [66, 67, 68, 69, 70, 87]. Today the parameterization of AM1* for the new d-block elements still continues. Comparisons of the AM1*, PM6 and PM5 methods for the transition metals are given in detail in the following sections.
11
INTRODUCTION Chapter 1
12
THEORY Chapter 2
2 THEORY
2.1 General Approximations
The most important goal of many approaches made in quantum chemistry is to solve the time- independent, non-relativistic Schrödinger equation [88] approximately.