ELECTRONIC STRUCTURE AND SPECTROSCOPY IN THE GAS...
Transcript of ELECTRONIC STRUCTURE AND SPECTROSCOPY IN THE GAS...
ELECTRONIC STRUCTURE AND SPECTROSCOPY IN THE GAS AND
CONDENSED PHASE: METHODOLOGY AND APPLICATIONS.
by
Vitalii Vanovschi
A Dissertation Presented to theFACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIAIn Partial Fulfillment of the
Requirements for the DegreeDOCTOR OF PHILOSOPHY
(CHEMISTRY)
January 2009
Copyright 2009 Vitalii Vanovschi
Acknowledgements
I’d like to thank my scientific adviser and mentor Prof. Anna I. Krylov. Her advice
and support were the critical success factors in the projects presented in this thesis. Her
enthusiasm toward the results I’ve obtained was highly encouraging. I appreciate the
mentoring on the number various topics Anna provided me with. Also, I would like to
thank Anna for the team-building activities she organized: hikes, trips, rock-climbing.
Thanks to the current and the former members of Anna’s group, especially Vadim
Mozhayskiy, Dr. Kadir Diri, Evgeny Epifanovsky, Lucasz Koziol, Dr. Sergey V. Levchenko
and Dr. Piotr A. Pieniazek for their help throught the projects.
I’d like to thank Prof. Curt Witting for the fascinating physical theories I’ve learned
in his class and for mentoring me during the course of studies.
I am grateful to our main collaborator on the effective fragment potential project,
Prof. Lyudmila Slipchenko for answering all the numerous questions I had about this
complicated method. Also I’d like to thank Prof. Mark S. Gordon for organizing a
welcome visit to Iowa State University where, working in the laboratory of the original
EFP inventors, I made a substantial progress on this project.
I’d like to thank Dr. Yihan Shao and Dr. Ryan Steele who helped me to figure out
the nuts and bolts of Q-Chem. Especially I’d like to thank Dr. Christopher Williams for
implementing the multipole integrals, required for the EFP electrostatics and polariza-
tion components.
ii
Table of Contents
Acknowledgements ii
List of Figures v
List of Tables viii
Abstract ix
1 Implementation of the Effective Fragment Potential method 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.3 Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141.4 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211.5 Exchange-repulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251.6 Implementation of EFP in Q-Chem . . . . . . . . . . . . . . . . . . . . 281.7 EFP application: Beyond benzene dimer . . . . . . . . . . . . . . . . . 371.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50Reference list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2 Structure, vibrational frequencies, ionization energies, and photoelectronspectrum of the para-benzyne radical anion 572.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572.2 Molecular orbital picture and the origin of vibronic interactions in the
para-benzyne anion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 602.3 Computational details . . . . . . . . . . . . . . . . . . . . . . . . . . . 652.4 Equilibrium structures, charge distributions, and theD2h→ C2v poten-
tial energy profiles at different levels of theory . . . . . . . . . . . . . . 672.5 Vibrational frequencies and the photoelectron spectrum of para-C6H
−4 . 77
2.6 DFT self-interaction error and vibronic interactions . . . . . . . . . . . 822.7 Calculation of electron affinities of p-benzyne . . . . . . . . . . . . . . 862.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90Reference list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
iii
3 Future work 963.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 963.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1003.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101Reference list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Bibliography 104
A GAMESS input for generating EFP parameters of benzene 117
B Geometries of benezene oligomers 119
C Converter of the EFP parameters 122
D Fragment-fragment electrostatic interaction 128
E Buckingham multipoles 130E.1 The conversion from the sperical to the Buckingham multipoles . . . . 130E.2 The conversion from the Cartesian to the Buckingham multipoles . . . . 132
F The Q-Chem input for EFP water cluster 134
G Q-Chem’s keywords affecting the EFP parameters computation 138
iv
List of Figures
1.1 Comparison of continuum (a) and discrete (b) models for a water moleculein the environment of other water molecules. . . . . . . . . . . . . . . 3
1.2 Uracil in aqueous solution. Active region (uracil) is shown with a ballsand sticks model. Effective fragments (water) are depicted by lines. . . 4
1.3 Distributed multipole expansion points for a water molecule. . . . . . . 10
1.4 Sketch of the charge penetration phenomenon for two water molecules. 13
1.5 The shape of the grid (monotonous gray area) for fitting the screeningparameters of a water molecule. . . . . . . . . . . . . . . . . . . . . . 15
1.6 Distributed polarizabilities for a water molecule. . . . . . . . . . . . . 16
1.7 Computing the induced dipoles through the two level self-consistentiterative procedure. HereΨ is the ab initio wavefunction,µ is a setof induced dipoles,FΨ is the force due to theab initio wavefunction andFµ is the force due to the induced dipoles. . . . . . . . . . . . . . . . . 19
1.8 Distributed dispersion points for a water molecule. . . . . . . . . . . . 23
1.9 The exchange-repulsion LMO centroids for a water molecule. . . . . . 27
1.10 Class diagram for the EFP energy part of the implementation. . . . . . 29
1.11 The internal structure of theFragmentParamsclass. . . . . . . . . . . 29
1.12 Representing the orientation with the Euler angles. xyz is the fixed sys-tem, XYZ is the rotated system, N is the intersection between the xy andXY planes called theline of nodes. Transformation between the xyz andXYZ systems is given byα rotation in the xy plane, followed byβ rota-tion in the new y’z’ plane (around the N axis), followed byγ rotation inthe new x”y” plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.13 The rotation matrix corresponding to the Euler anglesα, β, γ. . . . . . 31
v
1.14 The flow of control between Q-Chem and EFP energy module. . . . . . 32
1.15 The format for the$efp paramssection. . . . . . . . . . . . . . . . . . 35
1.16 The format for the$efp fragmentssection. . . . . . . . . . . . . . . . 35
1.17 Sandwitch (S), t-shaped (T) and parallel-displaced (PD) structures ofthe benzene dimer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.18 The structures of eight benzene trimers. . . . . . . . . . . . . . . . . . 45
1.19 Dependence of the non-additive energy share on the length of benzeneoligomers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.1 Frontier MOs and leading electronic configurations of the two lowestelectronic states atD2h andC2v structures. . . . . . . . . . . . . . . . . 62
2.2 The two lowest electronic states of C6H−4 can be described by: (i) EOM-
IP with the dianion reference; (ii) EOM-EA with the triplet neutral p-benzyne reference; (iii) EOM-SF with the quartet p-benzyne anion ref-erence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.3 D2h structures optimized at different levels of theory. . . . . . . . . . . 68
2.4 CCC angles, bonds lengths calculated by different coupled-cluster meth-ods and B3LYP. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.5 The distance between the radical-anion centers calculated by differentcoupled-cluster methods and B3LYP. . . . . . . . . . . . . . . . . . . . 70
2.6 The NBO charge at C1 (top panel) and dipole moments (bottom panel)at the optimizedC2v geometries. The ROHF-CCSD and ROHF-EA-CCSD values are calculated at the geometries optimized by the corre-sponding UHF-based method. . . . . . . . . . . . . . . . . . . . . . . 71
2.7 Energy differences between theC2v and D2h structures (6-311+G**basis). Upper panel: calculated using the UHF-CCSD optimized geome-tries. MCSCF and CASPT2N employ the MCSCF geometries. Lowerpanel: calculated at the respective optimized geometries. . . . . . . . . 73
2.8 The potential energy profiles along theD2h-C2v scans. At theD2h struc-ture, CCSD and CCSD(T) energies are calculated using symmetry-brokenand pure-symmetry UHF references (see section 2.3). . . . . . . . . . . 75
2.9 Displacements along the normal modes upon ionization (D2h structure). 80
vi
2.10 Top: Singlet (solid line) and triplet (dotted line) bands of the electronphotodetachment spectrum forD2h using the CCSD equilibrium geome-tries and frequencies of the anion. Bottom: the experimental (solid line)and the calculated (dotted line) spectra forD2h using the CCSD structureand normal modes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.11 The experimental (solid line) and the calculated (dotted line) spectra forC2v using the CCSD structure and normal modes. . . . . . . . . . . . . 83
2.12 The experimental (solid line) and the calculated (dotted line) spectra forD2h using the B3LYP structure and normal modes. . . . . . . . . . . . 84
2.13 Energy differences between theD2h andC2v minima (using the CCSDstructures) calculated using different density functionals. . . . . . . . . 87
2.14 Adiabatic electron affinities calculated by energy differences for the sin-glet (top panel) and triplet (bottom panel) state of the para-benzynediradical. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
vii
List of Tables
1.1 The geometric parameters and energies of the benzene dimers . . . . . 41
1.2 Dissociation energies of the benzene dimers at different levels of theory. 42
1.3 Total and many-body interaction energies of the benzene trimers (firstgroup). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.4 Total and many-body interaction energies of the benzene trimers (sec-ond group). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.5 Binding energies of benzene oligomers computed with EFP. . . . . . . 49
2.1 Energy differences between theC2v and D2h minima (eV) in the 6-311+G** basis set. Negative values correspond to theC2v minimumbeing lower. ZPEs are not included. . . . . . . . . . . . . . . . . . . . 74
2.2 Vibrational frequencies of the para-benzyne radical anion and the dirad-ical. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.3 Displacements in normal modes upon ionization,A ∗√
amu. . . . . . . 80
2.4 Electron affinities (eV) of the a3B1u and X1Ag states of the para-benzynediradical. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
viii
Abstract
Method developments and applications of the condensed phase and gas phase modeling
techniques are presented in two parts of this work.
In the first part, the theory of the effective fragment potential (EFP) method designed
for accurate modeling of the molecular properties of the extended systems is introduced.
The implementation of EFP developed in this work within the Q-Chem electronic struc-
ture package is discussed in details. The code is applied to the studies ofπ − π inter-
actions in benzene oligomers. The primary interest of these studies are the many-body
and total non-additiveπ − π interactions. A detailed comparison of EFP withab initio
theories is presented for the set of eight benzene trimers. Three-body intermolecular
interaction energies at the EFP level of theory are within 0.04 kcal/mol from the fullab
initio results. To elucidate the asymptomatic behavior of the total non-additive binding
energy, three types of linear benzene oligomers with 10, 20, 30 and 40 monomers are
modeled using our new EFP implementation. The result of this investigation is that with
the increase in the size of linear benzene clusters, the share of the total non-binding
energy approaches a constant and does not exceed 4% for the structures considered in
the study. This result indicates that there are no substantial non-local effects in the bind-
ing mechanism of linear benzene clusters.
In the second part, the equilibrium structure, vibrational frequencies, and ioniza-
tion energies of the para-benzyne radical anion are characterized by coupled-cluster
ix
and equation-of-motion methods. Vibronic interactions with the low-lying excited state
result in a flat potential energy surface along the coupling mode and even in lower-
symmetryC2v structures. Additional complications arise due to the Hartree-Fock insta-
bilities and near-instabilities. The magnitude of vibronic interactions was characterized
by geometrical parameters, charge localization patterns and energy differences between
theD2h andC2v structures. The observed trends suggest that theC2v minimum predicted
by several theoretical methods is an artifact of an incomplete correlation treatment. The
comparison between the calculated and experimental spectrum confirmedD2h struc-
ture of the anion, as well as accuracy of the coupled-cluster and spin-flip geometries,
frequencies and normal modes of the anion and the diradical. Density functional calcu-
lations (B3LYP) yielded only aD2h minimum, however, the quality of the structure and
vibrational frequencies is poor, as follows from the comparison to the high-level wave
function calculations and the calculated spectrum. The analysis of charge localization
patterns and the performance of different functionals revealed that B3LYP underesti-
mates the magnitude of vibronic interactions due to self-interaction error.
x
Chapter 1
Implementation of the Effective
Fragment Potential method
1.1 Introduction
Condensed phase encompasses the majority of matter on the Earth. Not surprisingly,
many molecular systems that are significant due to their biological, electrical or mechan-
ical properties also exist in the condensed phase. Properties of individual molecules in
such systems are influenced by the environment. Therefore, intermolecular interactions
play crucial role in chemical and physical properties of the condensed phase.
There are three distinct classes of methods for modeling the properties of the
extended systems at molecular level:ab initio (QM), classical (MM) and hybrid
(QM/MM) approaches.
The ab initio methods treat a molecule of interest together with the neighboring
molecules as one system at the same level ofab initio theory. The benefit of this
approach is that a high accuracy can be achieved, with a proper choice of theab ini-
tio method. However, computational costs of manyab initio methods scale steeply with
the size of the system [for instance, asN7 for CCSD(T)1], which makes the fullab initio
treatment of the environmental effects impractical.
New directions in the development of theab initio methods with more favorable
computational costs are:
1
• Linear scaling variants ofab initio methods2 are based on local and/or density
fitting (resolution of identity) approximations. At present, they are available for a
few ab initio methods (HF3, DFT4, MP25, CC6).
• Separation techniques such as semi-classical divide and conquer7 and frag-
ment molecular orbitals8 decompose the molecular system into fragments (e.g.,
molecules), which are described by individual and thus localized wavefunctions.
The wavefunctions for each fragment are computed in the field of the frozen wave-
functions of all other fragments until self-consistency is achieved.
• Plane-wave DFT represents the wavefunction in terms of the plane waves9. It
is suitable for systems with periodic boundary conditions (such as metals and
crystals) and features linear scaling of the computational effort.
On the other side of the theory spectrum, the whole molecular system can be mod-
eled using molecular mechanics, that is classically using empirical force fields10–17.
The advantage of this approach is its efficiency - very large systems with many
thousands of atoms can be modeled. However, molecular mechanics is incapable of
modeling chemical reactions or molecular properties depending on the electronic struc-
ture. Furthermore description of the intermolecular interactions is typically limited to
parametrized pairwise potentials and thus lacks accuracy.
In recent years, a number of hybrid methods combining the accuracy ofab initio
and the efficiency of molecular mechanics by treating different parts of the system with
different levels of theory has been developed18–21.
The hybrid methods can be subcategorized into continuum and discrete ones
(Fig. 1.1). The continuum methods consider the environment to be uniform, as the
name suggests. Its effects on an individual molecule is represented with a potential that
depends only on the shape of the cavity the molecule allocates and the bulk properties
2
of the matter. The simplicity of continuum models makes them computationally effi-
cient, but limit their accuracy. The discrete methods treat individual molecules of the
environment explicitly, for instance by including all the neighbor molecules within a
certain radius in the model. Unlike the fullyab initio methods, the molecules of the
environment are described classically, i.e., by a molecular mechanics (MM) force field.
The discrete methods are superior in the sense that their accuracy can be systematically
improved by (1) increasing the number of neighboring molecules included in the model
and by (2) using a higher level theory to describe them. Overall, the continuum models
are preferred when accurate description of the bulk is essential. For example, a shift in
the absorption spectrum of a molecule in a solvent can be well reproduced by a con-
tinuum model. They break down when the environment is non-uniform (e.g., enzyme
active center) or when relaxation is important due to the strong interaction between
solvent and solute. Such cases call for discrete methods. However, to describe bulk
behavior the discrete methods need to be coupled with a sampling technique (molecular
dynamics or Monte Carlo).
Figure 1.1: Comparison of continuum (a) and discrete (b) models for a water moleculein the environment of other water molecules.
3
Depending on the way QM and MM parts are integrated with each other, the dis-
crete QM/MM methods can be categorized as mechanical embedding and electrostatics
embedding18. Mechanical embedding computes the MM parameters from the QM wave-
function and then treats the whole system classically. Although mechanical embedding
is simple and computationally cheap, the large drawback is that the perturbation of the
QM wavefunction due to MM part is ignored. Electrical embedding explicitly accounts
for the presence of the MM part by including the perturbative potential from MM into
the Hamiltonian of the QM system. Thus, it provides better description of the electronic
structure of the QM part, at the price of the increased computational complexity.
One of the promising hybrid methods based on the electrostatic embedding is the
effective fragment potential (EFP) approach22–33.
In EFP the molecular system is decomposed into an active region and a number of
effective fragments. An example of such decomposition for uracil (active region) in
water (effective fragments) is demonstrated in Fig. 1.2.
Figure 1.2: Uracil in aqueous solution. Active region (uracil) is shown with a balls andsticks model. Effective fragments (water) are depicted by lines.
The active region is a part of the molecular system that requires accurate description,
for instance, because it participates in a chemical reaction or its electronic properties are
4
of interest. Therefore, the active region is modeled with anab initio method. The effec-
tive fragments are the spectator molecules and thus do not require high accuracy for their
internal description. However, since the interaction between the active region and the
effective fragments affects the properties of the active region, this interaction should also
be modeled with reasonable accuracy. An effective fragment in EFP is described semi-
classically by a set of multipoles (with screening coefficients), polarizability points,
dispersion points and exchange-repulsion parameters.
The interaction between the effective fragments and the active region is accounted
for by adding a one-electron term to the Hamiltonian of the active region:
H ′ = H + 〈p |V | q〉 , (1.1)
whereV is the potential due to the effective fragments.
The expression for the inter-fragment interaction energy is derived by applying
perturbation theory to the wavefunctions of two interacting molecules, as detailed in
Ref. 53. Within this approach, the total energy for fragments A and B is:
Etotal = E0 + E1 + E2 + ...
E0 = EA + EB
E1 = 〈00 |T | 00〉
E2 = −∑mn
〈00 |T |mn〉 〈mn |T | 00〉E0
mn − E000
,
(1.2)
whereEA andEB are energies of the individual isolated fragments,En are the pertur-
bative corrections,T is the electrostatics operator,E0mn is the energy of the excited state
mn for the two non-interacting fragments.
5
The first order perturbative correctionE1 is the electrostatic interaction. In EFP,
it is described by a distributed multipole expansion (section 1.2). The second order
perturbative correction consists of polarization and dispersion terms. The polarization
component of EFP is accounted for by distributed polarizabilities (section 1.3). The
dispersion is represented similarly by a set of distributed dispersion points (section 1.4).
These terms are derived from the exactab initio expressions and, therefore, their accu-
racy can be systematically improved. The higher order terms in perturbative expansion
are expected to have smaller values and are neglected in the current EFP formalism.
One of the reasons of the EFP success29 is that some of the higher order terms are
believed to cancel out each other, thus omitting them does not affect the total inter-
action energy substantially. For example, charge-transfer and exchange-induction are
similar in magnitude, but have the opposite signs and thus cancel out each other. The
exchange-dispersion partially cancels out with higher order dispersion terms (such as
induced dipole-induced quadrupole) that are also omitted in the EFP formalism29.
Perturbative treatment of intermolecular interactions described above works well for
large distances, but breaks down when molecules are close to each other and their elec-
tronic densities overlap. To account for this effect, EFP includes a charge-penetration
correction (section 1.2) to the electrostatics and the exchange-repulsion component (sec-
tion 1.5) accounting for the Pauli exclusion principle. Both terms are also derived from
the exactab initio expressions. Thus, EFP intermolecular interaction consists of four
components: electrostatics (with charge-penetration correction), polarization, disper-
sion and exchange-repulsion.
The interaction energy between the effective fragments is computed as:
Eef−ef = Eelec + Epol + Edisp + Eex−rep (1.3)
6
Since the active region is described by anab initio method, the electrostatics and
polarization interactions with the effective fragments are accounted for by additional
one-electron terms in the Hamiltonian:
H ′ = H +⟨p∣∣V elec + V pol
∣∣ q⟩ (1.4)
In the present EFP model, dispersion and exchange-repulsion interactions between
active region and the effective fragments are treated similarly to the fragment-fragment
interactions as additive corrections to the total energy.
One of the advantages of EFP is that all of its terms has been derived from the exact
ab initio expressions. Thus, the parameters necessary to describe an effective fragment
can be obtained from the wavefunction of an isolated molecule. The parameters are
computed only once for every fragment type and then can be reused for all consequent
computations. This makes EFP method very efficient. Its computational cost isO(N2)
for the system withN effective fragments, plus the cost of theab initio computation for
the active region, which does not depend on the number of fragments.
Another advantage of EFP is that, unlike standard QM/MM approaches, it is a polar-
ized force field. Therefore, the effective fragments not only polarize the active region,
but also are polarized by it (polarized embedding), as well as by each other. Therefore,
EFP describes polarization effects more accurately than standard QM/MM approaches.
The third advantage is that the accuracy of EFP can be systematically improved
by introducing higher order terms and cross terms in the perturbative expansion (e.g.
exchange-dispersion), as well as higher order terms in the expansion of the individual
components (e.g. hexapole-hexapole term for electrostatics, or induced dipole-induced
quadrupole term for dispersion).
7
The first implementation of EFP appeared in the GAMESS34 quantum chemistry
package, which is developed by the original authors of the EFP method22. By now
EFP has been successfully applied for treating the environmental effects in a number
of molecular systems35–41. In addition, due to the accurate description of the inter-
fragment interactions, EFP has been successfully used to study the relative stabilities of
the molecular clusters and the dissociation energy profiles27,28,42–46.
The primary goal of this work was to implement the EFP method in Q-Chem47. The
benefits of such implementation are: (1) the extended availability of the EFP method, (2)
the ability to generate EFP fragments parameters with a higher level methods available
in Q-Chem (such as CCSD) and (3) creating the base for future development of EFP in
which the active region can be modeled with a correlatedab initio theorya.
The second goal was to apply the developed EFP implementation to the benzene
clusters. These systems are interesting due to theirπ-π stacking interactions, which are
similar in nature to those in DNA48.
The individual components of EFP are described in details in the four following
sections. Section 1.6 describes the developed implementation of EFP in Q-Chem. Its
application to benzene trimers and higher oligomers is presented in section 1.7. The last
section of the chapter presents summary of the results and conclusion.
aThe present GAMESS implementation only enables a non-correlated description of the active region(HF or DFT).
8
1.2 Electrostatics
Electrostatic component of the EFP energy accounts for Coulomb interactions. For
molecular systems with hydrogen bonds or for strongly polar molecules, electrostatics
has the leading contribution to total EFP energy49.
EFP electrostatics was devised to provide an accurate, yet compact and computa-
tionally efficient description of the molecular electrostatic properties22. Buckingham49
has shown that the Coulomb potential of a molecule can be expressed using multipole
expansion at a single spacial pointk:
V eleck (x) =
qk
rkx
+
x,y,z∑a
µkaFa(rkx) +
1
3
x,y,z∑a,b
ΘkabFab(rkx)−
1
15
x,y,z∑a,b,c
ΩkabcFabc(rkx), (1.5)
whereq, µ, Θ, andΩ are net charge, dipole, quadrupole, and octopole, respectively, and
Fa, Fab, andFabc are the electric field, field gradient, and field second derivatives, due
to the effective fragment. This expression converges to the exact electrostatic poten-
tial when an infinite number of terms are included. However, it is known to converge
slowly50 and thus requires a large number of terms to achieve a reasonable accuracy,
even if such expansions are assigned to the individual atoms in the molecule, as in Mul-
liken population analysis51. That makes a single point or atomic based representations
of the molecular electrostatic potential impractical.
In his work on distributed multipoles analysis50,52–54, Stone has shown that a better
description of the Coulomb potential can be obtained by using many points as centers
of the multipole expansions. In fact, for a system with a wavefunction expressed in the
basis ofN gaussians, the Coulomb potential can be described exactly with a finite num-
ber of multipoles located atN2 points corresponding to the gaussian products centers55.
However, this representation becomes inefficient for systems with a large basis.
9
Numerical tests conducted on a number of systems have demonstrated that an accu-
rate representation of the electrostatic potential can be achieved by using multipoles
expansion around atomic centers and bond midpoints (that is, the points known to have
high electronic density) and truncating multipoles expansions after the octopoles50,52.
This representation has been adopted by EFP22,25.
Thus, electrostatic potential of an effective fragment is specified by charges, dipoles,
quadrupoles and octopoles placed at atomic centers and bond midpoints of a molecule.
An example of such expansion for a water molecule is shown in Fig. 1.3.
Figure 1.3: Distributed multipole expansion points for a water molecule.
The energy of electrostatic interaction for a system of effective fragments can be
expressed as:
Eelec =1
2
∑A
∑B 6=A
(∑k∈A
∑l∈B
Eeleckl +
∑I∈A
∑J∈B
EelecIJ +
∑I∈A
∑l∈B
EelecIl
), (1.6)
whereA andB are the fragments,k andl are the multipole expansion points,I andJ
are the nuclei.
10
The electrostatic interaction between two expansion points is computed using the
following equation22:
Eeleckl = Ech−ch
kl + Ech−dipkl + Ech−quad
kl +
Ech−octkl + Edip−dip
kl + Edip−quadkl + Equad−quad
kl
(1.7)
The specific equations for charge-charge (D.1), charge-dipole (D.2), charge-quadrupole
(D.3), charge-octopole (D.4), dipole-dipole (D.5), dipole-quadrupole (D.6) and
quadrupole-quadrupole (D.7) are provided in Appendix D.
Nuclei-multipole and nuclei-nuclei terms are given by equations (1.8) and (1.9)
respectively22.
EelecIl = Enuc−ch
Il + Enuc−dipIl + Enuc−quad
Il + Enuc−octIl (1.8)
EelecIJ =
ZIZJ
RIJ
(1.9)
Equations for nuclei-charge (D.8), nuclei-dipole (D.9), nuclei-quadrupole (D.10) and
nuclei-octopole (D.11) terms can be found in Appendix D.
Electrostatic interaction between an effective fragment and theab initio part is
described by introducing perturbation toab initio Hamiltonian:
H = H0 + V (1.10)
The perturbation is a one-electron term computed as a sum of contributions from all the
expansion points and nuclei of all the effective fragments:
V =∑
p
∑q
∑A
(∑k∈A
dpq
⟨p∣∣V elec
k
∣∣ q⟩+∑I∈A
⟨p
∣∣∣∣ZI
R
∣∣∣∣ q⟩)
, (1.11)
11
whereA is an effective fragment,I is a nucleus,p andq are atomic orbitals of theab
initio part,ZI is nucleus charge,dpq is an element of the atomic density matrix andV eleck
is electrostatic potential from an expansion point.
A one-electron contribution to the Hamiltonian from an individual expansion point
consists of 4 terms, each of which originates from the electrostatic potential of the cor-
responding multipole22:
⟨p|V elec
k |q⟩
=
⟨p
∣∣∣∣qk
R
∣∣∣∣ q⟩+
⟨p
∣∣∣∣∑x,y,za µk
aa
R3
∣∣∣∣ q⟩+
⟨p
∣∣∣∣∣∑x,y,z
a,b Θkab(3ab−R2δab)
3R5
∣∣∣∣∣ q⟩
+
⟨p
∣∣∣∣∣∑x,y,z
a,b,c Ωkabc(5abc−R2(aδbc + bδac + cδab))
5R7
∣∣∣∣∣ q⟩
,
(1.12)
whereR, a, b andc are the distance and distance components between the electron and
the expansion pointk.
Electrostatics parameters
The parametersq, µ, Θ, Ω of the effective fragment electrostatics need to be computed
only once for every fragment type. These multipoles are obtained from anab initio
electronic density of the individual molecule using DMA (distributed multipoles analy-
sis) procedure described by Stone50,52. First, a molecular wavefunction expressed in a
basis ofN gaussians is computed with anab initio method. Then spherical multipoles
expansions are calculated at the centers of the gaussian basis function products. These
expansions are finite and include multipoles up toL1 + L2 order whereL1 andL2 are
angular moments of the two gaussians. At the next step, these multipoles are trans-
lated to the EFP expansion points (atomic centers and bond midpoints) as described
in Ref. 50 and truncated after the4th term. Thus obtained set of spherical multipoles
12
is then converted to traceless Buckingham multipoles using the equations provided in
Appendix E.1.
Charge penetration
Model of EFP electrostatics described above works well when electronic densities of the
effective fragments do not overlap. However, for many molecular systems the overlap
is substantial and is known as charge-penetration effect. A sketch demonstrating charge
penetration for two water molecules is shown in Fig. 1.4
Figure 1.4: Sketch of the charge penetration phenomenon for two water molecules.
The magnitude of this effect is commonly around 15% of the total electrostatic
energy and for some systems is as large as 200%27. It was demonstrated in Ref. 24 that
charge penetration can be efficiently accounted for by introducing a screening function
as a prefactor to the Coulomb potential:
V elecµk (x) →
(1− e−αµkrx−µk
)V elec
µk (1.13)
13
This screening is applied only to the potential originating from charges representing
electronic density, and not to the potential of higher multipoles, which represents elec-
tronic density variations27. Therefore only the charge-charge term is modified to account
for the charge-penetration24:
Ech−chkl =
(1− (1 + αkRkl
2)e−αkRkl
)qkql
rklif αk = αl(
1− α2l
α2l−α2
ke−αkRkl − α2
k
α2k−α2
le−αlRkl
)qkql
rklif αk 6= αl
(1.14)
Electrostatics screening parameters
Parameters a and b of the screening function are computed using a fitting procedure
described in Ref. 24 once for every type of the fragment. This procedure optimizes the
screening parameters so that the sum of differences between theab initio electrostatic
potential and the EFP electrostatic potential with screening is minimized:
∆ =∑
p
[V Elec
ab initio(p)− V ElecEFP (p)
], (1.15)
where the sum is taken over all the pointsp of the grid. The grid is constructed such that
it includes the points with substantial electronic density and excludes the points close to
nuclei and bond midpoints, which are inaccessible to multipoles of another fragments.
A projection of such grid for a water molecule is shown in Fig. 1.5.
1.3 Polarization
Polarization accounts for the intramolecular charge redistribution under the influence of
an external electric field. It is the major component of many-body interactions, which
14
Figure 1.5: The shape of the grid (monotonous gray area) for fitting the screening param-eters of a water molecule.
are important for proper description of cooperative molecular behavior. Since polariza-
tion cannot be parametrized by two-body potentials, it is often disregarded in molecular
mechanics models. However, for hydrogen bonded systems, up to 20% of the total
intermolecular interaction energy comes from polarization22.
The exact expression for the polarization energy appears as the second-order term of
a long-range perturbation theory53:
Epol = −∑n6=0
〈00|V |0n〉 〈0n|V |00〉En − E0
, (1.16)
wheren is the excited electronic state andV is the perturbing electrostatic potential.
Expanding electrostatic potentialV in(
1R
)nseries in equation (1.16) and then com-
bining terms with the same electric field derivatives yields the following equation:
Epol = −1
2αabFaFb −
1
6βabcFaFbFc −
1
3Aa,bcFaFbc
−1
6Bab,dcFaFbFcd −
1
6Cab,dcFabFcd + ...,
(1.17)
15
whereFa andFab are electric field derivatives,α is the dipole polarizability tensor,β is
the dipole hyperpolarizability,A, B andC are quadrupole polarizability tensors.
In order to achieve a better convergence, EFP uses distributed polarizabilities placed
at the centers of valence LMOs. This allows one to truncate the expansion (1.17) after
the first term and still retain a reasonable accuracy22. It should be noted that unlike the
total molecular polarizability tensor, which is symmetric, the distributed polarizability
tensors are asymmetric and thus all of 9 tensor components are required. Polarizability
points for a water molecule are shown in Fig. 1.6.
Figure 1.6: Distributed polarizabilities for a water molecule.
Let us discuss how polarization energy is computed for an individual effective frag-
ment in an external force field. First, induced dipoles for every polarizability point are
computed:
µk = αkFk, (1.18)
whereµk, αk andFk are the induced dipole, the polarizability tensor and the external
force field at point k.
16
Then induced dipoles are used to compute the polarization energy:
Epol =∑
k
µkFk (1.19)
Now, let us consider a system with many effective fragments and anab initio part.
In this case, the electric field acting on an effective fragment depends on: multipoles
and nuclei of other effective fragments, induced dipoles of other effective fragments and
electronic density and nuclei of theab initio part:
Fk =∑B 6=A
(∑i∈B
Fmulti (k) +
∑I∈B
F nucI (k) +
∑j∈B
F indj (k)
)
+F ab initio elec(k) + F ab initio nuc(k),
(1.20)
whereFk is the total electric field at polarizability pointk of the effective fragmentA, j
is a polarizability point,i is a multipole expansion point,I is a nucleus,Fmulti (k) is the
field due to the multipolei, F nucI (k) is the field from the nucleusI, F ind
j (k) is the field
due to the induced dipolej, F ab initio elec(k) andF ab initio nuc(k) are the forces from the
electron density and nuclei of theab initio part.
The dependence of the total electric field on other induced dipoles andab initio
density gives rise to two complications in computing the polarization energy discussed
below.
The first complication arises from the fact that an induced dipole depends on the
values of all other induced dipoles in all other fragments in according with equations
(1.18) and (1.20). Therefore, they must be computed self-consistently.
17
Another complication originates from the dependence of the induced dipoles on the
ab initio electron density, which, in turn, is affected by the field created by these induced
dipoles through a one electron contribution to the Hamiltonian:
H ind =∑
A
∑k∈A
∑x,y,za (µa
k + µak)a
R3, (1.21)
whereR, a, b andc are the distance and distance components between the electron and
the polarizability pointk, µak andµa
k are components of the induced dipole and conju-
gated induced dipole, respectively. Thus, when theab initio part is present, conjugated
induced dipoles should also be computed:
µk = αTk Fk, (1.22)
whereαTk andFk are the transposed polarizability tensor and the external force field at
the polarization point k. In the following discussion it is implied that the conjugated
induced dipoles are always computed together with the regular induced dipoles.
As a result of these self-consistency requirements, the total polarization energy is
computed with a two level iterative procedure illustrated in Fig. 1.7. The objective of the
higher level is to converge the wavefunction. The lower level is tasked with converging
the induced dipoles for a given fixed wavefunction.
In the beginning, the induced dipoles are initialized to 0 and the electronic density is
obtained through a guess procedure (such as a generalized Wolfsberg-Helmholtz proce-
dure56).
At every iteration of the higher level, the wavefunction is updated based on the values
of the induced dipoles on the previous step. Then the force from this new wavefunction
acting on the polarizability points is recomputed. The convergence criteria is that the
18
Is μ converged?
μ = 0Ψ = guess()
Compute Ψ
Compute FΨ
Compute Fμ
Compute μ
No
Is Ψ converged?
Yes
Exit
No Yes
Figure 1.7: Computing the induced dipoles through the two level self-consistent iterativeprocedure. HereΨ is theab initio wavefunction,µ is a set of induced dipoles,FΨ is theforce due to theab initio wavefunction andFµ is the force due to the induced dipoles.
wavefunction parameters (molecular orbital coefficients) are within a predefined range
from their values on the previous iteration.
At the lower level, the values of the induced dipoles are computed at every iteration
based on the electric force from theab initio part (which does not change during the
lower-level iterations), as well as the forces from multipoles, nuclei and the force from
the induced dipoles of the previous lower-level iteration. The convergence criteria is that
the difference between the new induced dipoles and those from the previous iteration is
within a predefined threshold.
Thus, when the lower-level iterative procedure exits the induced dipoles are self-
consistent and are consistent with a fixedab initio wavefunction. Convergence of the
19
higher-level iterative procedure results in the induced dipoles and theab initio wave-
function that are self-consistent and consistent with each other. If the molecular system
does not contain theab initio part, then only the lower-level procedure is required.
Once the values of the induced dipoles are obtained, the total energy of the system
can be computed as22:
E = Eother − 1
2
∑A
∑k∈A
µk(Fmultk + F nuc
k ) +1
2
∑A
∑k∈A
(µk + µk)Fab initio eleck , (1.23)
whereEother is the sum of energies of theab initio part and other components of EFP,
A is an effective fragment,k is a polarizability point,µk andµk are induced dipoles and
conjugated induced dipoles,Fmultk is the force due to the multipole expansion points of
other fragments,F nuck is the force due to the all other nuclei,F ab initio elec
k is the force due
to theab initio electronic density (F ab initio eleck =
⟨Ψ∣∣∣f elec
k
∣∣∣Ψ⟩).
In equation (1.23), the second term is the total polarization energy of all the effective
fragments. However, the second and the third term do not add up to the total polarization
energy because a part of the later is implicitly included in theab initio energy (due to
the effect of the induced dipoles on the wavefunction).
Computational cost of the polarization component consists of the cost of computing
the induced dipoles and the energy. The latter is a trivial step with the cost ofO(N2) for
N effective fragments. Thus, the former defines the overall computational cost.
At the higher level, the number of iterationsh is similar to that for theab initio
system without EFP, while at the lower level the convergence is achieved within bounded
number of iterations. Every lower-level iteration scales asO(N2), thus, the overall
computational cost isO(hN2).
If system does not have theab initio part, then only the lower-level iterative proce-
dure is required and computational cost isO(N2).
20
Polarization parameters
The polarization parameters in the EFP model are LMOs and polarizability tensors.
Both can be obtained from anab initio calculation once for every type of the fragment
and then reused in all consequent calculations. LMOs are obtained with the Boys local-
ization procedure based on theab initio electronic density57,58.
Polarizabilities are obtained based on the definition (1.24) as a derivative of the
dipole moment (µ) w.r.t. electric field (F ) using the finite differences method.
αab =∂µa
∂Fb
, (1.24)
In this approach, a dipole moment of an LMO is first computed without an electric field.
Then, by means of 3 additionalab initio computations, the LMO and its dipole moment
are computed in the presence of a small electric field acting along one of the coordinate
axis. Based on LMO dipole moments, the polarizability tensor can be computed as:
αab =µa(Fb)− µa(0)
Fb
(1.25)
1.4 Dispersion
For many molecular systems dispersion is the main attractive component of the inter-
molecular interaction energy. An accurate description of dispersion interaction is a diffi-
cult task even forab initio methods — inclusion of the electron correlation at high level
is required. An additional challenge for EFP is that its dispersion description must be
computationally efficient.
21
The exact expression (1.26) for the dispersion energy appears in the second order
term of a long-range perturbation theory.
Edisp = −∑n6=0
∑m6=0
〈00|V |mn〉 〈mn|V |00〉EA
m + EBn − EA
0 − EB0
, (1.26)
whereA andB are two interaction molecules,m andn are their excited electronic states
andV is the perturbing electrostatic potential.
By expanding electrostatic potentialV in(
1R
)nseries dispersion energy can be
expressed in the so called London series:
Edisp =C6
R6+
C8
R8+
C10
R10+ ..., (1.27)
where theCn coefficients correspond to induced dipole—induced dipole (C6), induced
dipole—induced quadrupole (C8), induced quadrupole—induced quadrupole and
induced dipole—induced octopole (C10) etc. interactions, Thus, dispersion energy can
be alternatively expressed as a sum:
Edisp = Eind dip−ind dip + Eind dip−ind quad + Eind quad−ind quad + Eind dip−ind oct + ...
(1.28)
A better convergence can be obtained by using a distributed dispersion model with
expansion points located at localized molecular orbitals centroids59. Expansion points
for a water molecule are demonstrated in Fig. 1.8.
When using distributed dispersion, a reasonable accuracy can be achieved by trun-
cating the expansion after the first term and approximating the rest as1/3 of the this
22
Figure 1.8: Distributed dispersion points for a water molecule.
term59. Induced dipole-induced dipole interaction energy between two induced dipoles
k and l can be computed by using the following equation:
Eind dip−ind dipkl =
x,y,z∑abcd
T klabT
klcd
∫ ∞
0
αkac(iν)αl
bd(iν)dν, (1.29)
whereαkac andαl
bd are dynamic polarizability tensors,iν is the the imaginary frequency,
T klab andT kl
cd are the electrostatic tensors of the second rank, i.e.,T klab =
3ab−R2klδab
R5kl
.
In general, the interaction energy described by (1.29) is anisotropic, distance depen-
dent and computationally expensive26. The approximation used by EFP includes only
the trace of polarizability tensor, which is anisotropic and distance independent.
The total dispersion energy for a system of effective fragments is:
Edisp =4
3
∑A
∑B
∑k∈A
∑l∈B
Ckl6
R6kl
, (1.30)
23
whereA andB are effective fragments,k andl are LMOs,Ckl6 is intermolecular disper-
sion coefficient, andR6kl is the distance between two LMO centroids.
Intermolecular dispersion coefficients are computed using the 12 point Gauss-
Legendre integration formula:
Ckl6 =
12∑i=1
wi2ν0
(1− ti)2αk(iνi)α
l(iνi), (1.31)
whereαk(iνi) andαl(iνi) are1/3 of the traces of the corresponding polarizability ten-
sors.
Dispersion damping
For small distances between effective fragments dispersion coefficients must be cor-
rected to take into account the charge penetration effect. In the original EFP model, the
Tang—Toennies formula60 with parameterb = 1.5 is used:
Ckl6 →
(1− e−bR
∞∑k=0
(bR)k
k!
)Ckl
6 (1.32)
However, recent studies revealed that this formula over-damps the dispersion and a more
appropriate value forb lies between2 and3 and is system dependent29. As an alternative,
Ref. 29 suggested to use a damping function dependent on the electronic density overlap
Skl between LMOsk andl:
Ckl6 →
(1− |Skl|2(1− 2ln|Skl|+ 2ln2|Skl|)
)Ckl
6 (1.33)
24
Dispersion parameters
As follows from equations (1.30) and (1.31), the dispersion parameters for an effective
fragment are the coordinates of LMO centroids and sets of dynamic polarizability ten-
sor traces for 12 frequencies for each LMO. These parameters can be computed once
for every type of an effective fragment and then reused in EFP calculations. Dynamic
polarizabilities can be computed using dynamic time-dependent Hartree-Fock (TDHF)
or dynamic time-dependent density functional theory (TDDFT) as described in Ref. 26.
1.5 Exchange-repulsion
Exchange-repulsion originates from the Pauli exclusion principle, which states that for
two identical fermions the wavefunction must be anti-symmetric. Inab initio methods
such as Hartree-Fock and DFT, it is accounted for by the exchange operator. In classical
force fields, exchange-repulsion is introduced as a positive (repulsive) term, e.g.,(
1r
)12in the Lennard-Jones potential. However, since the Pauli exclusion principle is intrin-
sically quantum, its classical descriptions are not accurate. EFP uses a wavefunction-
based formalism to account for the electron exchange. Thus exchange-repulsion is the
only non-classical component of EFP and the only one that is repulsive.
The EFP exchange-repulsion energy term has been derived from the exact equation
for the exchange-repulsion energy23,61:
Eexch =
⟨ΨAΨB|AHAB|ΨAΨB|
⟩⟨ΨAΨB|AΨAΨB
⟩ − 〈ΨAΨA|VAB|ΨBΨB〉 − EA − EB, (1.34)
25
whereA andB are two molecules described byΨA andΨB wavefunctions,VAB is the
intermolecular Coulomb interaction, andA is the antisymmetrization operator:
A = P0 − P1 + P2 − P3 + ..., (1.35)
wherePi is a permutation ofi electron pairs consisting of one electron from each
molecule.
Truncating sequence (1.35) after the second term, applying an infinite basis set
approximation and a spherical gaussian overlap approximation as shown in Ref. 61
leads to the expression for the exchange-repulsion energy in terms of localized molecu-
lar orbitals:
Eexch =1
2
∑A
∑B 6=A
∑i∈A
∑j∈B
Eexchij (1.36)
Eexchij = −4
√−2ln|Sij|
π
S2ij
Rij
−2Sij
(∑k∈A
FAikSkj +
∑l∈B
FBjl Sil − 2Tij
)
+2S2ij
(∑J∈B
−ZJR−1iJ + 2
∑l∈B
R−1il
∑I∈A
−ZIR−1Ij + 2
∑k∈A
R−1kj −R−1
ij
),
(1.37)
whereA andB are the effective fragments,i, j, k andl are LMOs,I andJ are nuclei,
S andT are intermolecular overlap and kinetic energy integrals,F is the Fock matrix
element.
The positions of LMO centroids for a water molecule are shown in Fig. 1.9.
Formula for theEexchij involves intermolecular overlap and kinetic energy integrals,
which are expensive to compute. In addition, since equation (1.37) is derived within an
26
Figure 1.9: The exchange-repulsion LMO centroids for a water molecule.
infinite basis set approximation, it requires a reasonably large basis set to be accurate.
These factors make exchange-repulsion the most computationally expensive part of the
EFP energy calculations of moderately sized systems.
Large systems require additional considerations. Equation (1.36) contains a sum
over all the fragment pairs and its computational cost formally scales asO(N2) with
the number of effective fragmentsN . However, exchange-repulsion is a short range
interaction —- the overlap and kinetic energy integrals of LMOs vanish very fast with
the interfragment distance. Therefore, by employing a distance based screening, the
number of overlap and kinetic energy integrals scales asO(N). As a result, for large
systems exchange-repulsion becomes less computationally expensive than long range
components of EFP (such as Coulomb interactions).
Exchange-repulsion parameters
The parameters required for equation (1.37) are the Fock matrix elements and the LMOs
(orbital coefficients, basis and centroids). They can be computed once for every type of
27
the fragment by performing a Hartree-Fock calculation on the molecule followed by an
an orbital localization procedure (such as Boys LMO)23.
1.6 Implementation of EFP in Q-Chem
This section describes the new implementation of the EFP method in the Q-Chem elec-
tronic structure package.
As detailed in section 1.1, within EFP formalism the molecular system is decom-
posed into an active region and a number of effective fragments. Since the active region
is treated by anab initio method, EFP requires an interface with a general-purpose quan-
tum chemistry package. Our implementation was developed using C++62 as a module
within the state of-the-art quantum chemistry package Q-Chem47. The standard template
library (STL) was extensively used throughout the implementation.
The developed code consists of the two independent parts: EFP energy and EFP
parameters. The former is responsible for computing energy of the molecular system,
whereas the latter allows one to obtain effective fragment parameters.
Computing EFP energy
The EFP energy part of the implementation allows one to compute total energy of a
system comprised of an optional active region and a number of effective fragments.
The total energy includes contributions from all the EFP components described in the
four previous sections: electrostatics (with charge-penetration correction), polarization,
dispersion, exchange-repulsion.
A simplified high-level class diagram for the EFP energy code is shown in Fig. 1.10
(a dashed arrow represents a dependence). The main classEFP is responsible for man-
aging the interactions between other classes and providing the interface to the rest of
28
the Q-Chem package. ClassesElectrostatics, Polarization, DispersionandExchange-
repulsionencapsulate the functionality required for the corresponding individual energy
components. ClassPolarizationhas the dependence on the classElectrostaticsbecause
induced dipoles depend on the field created by multipoles. The responsibility of the
ConfigReaderclass is reading the effective fragment parameters and configuration of
the molecular system, and storing them in the internal representation using classesFrag-
mentParamsandFragment.
Dispersion
EFP
Electrostatics
Polarization ExchangeRepulsion ConfigReader
FragmentParams Fragment
Figure 1.10: Class diagram for the EFP energy part of the implementation.
FragmentParams
std::vector< double > er_fock_matrix
std::vector< PolarizationPoint >polarizabilities
std::vector< XYZPoint >er_lmos
std::vector< std::vector< double > > er_wavefunction
std::vector< DispersionPoint >dispersions
std::string
er_basis
std::vector< EFPAtom >
atoms
std::vector< MultipolePoint >
multipoles
Figure 1.11: The internal structure of theFragmentParamsclass.
29
The internal structure forFragmentParamsclass is shown in Fig. 1.11. The labels
above the arrows indicate the name of the data member. This class encapsulates all
the parameters for a given EFP fragment type: atoms, multipoles, dispersion points,
polarizabilities and parameters for exchange-repulsion (which start with erprefix).
The classFragmentrepresents an individual effective fragment in the molecular sys-
tem. It stores the fragment type, its position and orientation. The position is specified
in terms of translation in Cartesian coordinates (x, y, z) relative to the origin of theab
initio system. The orientation is specified by the Euler angles (α, β, γ) in terms of the
3 consequent rotations, as explained in Fig. 1.12. The rotation matrix corresponding to
these angles is given in Fig. 1.13.
N
x
y
z
Z
X
Y
α
β
γ
Figure 1.12: Representing the orientation with the Euler angles. xyz is the fixed system,XYZ is the rotated system, N is the intersection between the xy and XY planes calledthe line of nodes. Transformation between the xyz and XYZ systems is given byαrotation in the xy plane, followed byβ rotation in the new y’z’ plane (around the Naxis), followed byγ rotation in the new x”y” plane.
30
R = cosα cosγ − sinα cosβ sinγ −cosα sinγ − sinα cosβ cosγ sinβ sinα
sinα cosγ + cosα cosβ sinγ −sinα sinγ + cosα cosβ cosγ −sinβ cosα
sinβ sinγ sinβ cosγ cosβ
Figure 1.13: The rotation matrix corresponding to the Euler anglesα, β, γ.
The positionx′i, y′i, z
′i of an atomi of an effective fragment can be computed as:
x′i
y′i
z′i
= R
xi
yi
zi
+
x
y
z
, (1.38)
wherexi, yi, zi are the coordinates of the atom in the fragment frame,R is the rotation
matrix defined in Fig. 1.13, andx, y, z are the fragment coordinates.
The flow of control between Q-Chem and the EFP module is presented in Fig. 1.14.
TheReadParamsandReadFragmentscalls instruct the EFP module to read the defini-
tions of the fragment types and the molecular system configuration from the Q-Chem
standard input (discussed in the next subsection). TheInit call checks the effective
fragments parameters and the configuration for consistency, creates and initializes the
instances ofElectrostatics, Polarization, DispersionandExchange-Repulsionclasses.
TheUpdateRepresentationmethod is called every time when the geometry of the molec-
ular system changes. It performs rotation and translation of atoms, multipoles, polariza-
tion tensors, dispersion points and LMOs in accordance with the coordinates and ori-
entation for each individual fragment. TheNuclearEnergyfunction computes the EFP
energy components, which are independent from the wavefunction ofab initio system
(inter-fragment electrostatics, dispersion and exchange-repulsion).
31
If the active region is present, then for every iteration of the SCF procedure Q-Chem
calls theHamiltonianmethod and theWFDependentEnergyfunction. TheHamiltonian
method updates the one-electron term of the active region Hamiltonian in accrodance
with the equations (1.10) and (1.21). After updating the wavefunction accordingly, Q-
Chem callsWFDependentEnergyto compute the wavefunction-dependent part of the
EFP energy (polarization).WFDependentEnergytakes the electronic density of the
active region as a argument. If the active region is absent, Q-Chem callsWFDepen-
dentEnergyonly once with a zero electronic density as an argument.
Finally, thePrintSummarymethod is called to print the individual energy compo-
nents for the inter-fragment EFP energy.
Q-Chem EFP
ReadParams()
ReadFragments()
Init()
UpdateRepresentation()
NuclearEnergy()
Hamiltonian()
WFDependentEnergy()
energy
energy
for every SCF cycle
PrintSummary()
hamiltonian
for every new geometry
Figure 1.14: The flow of control between Q-Chem and EFP energy module.
32
The correctness of the EFP energy implementation in Q-Chem was verified by
performing the energy computation on a few dozens of examples and comparing the
results with the GAMESS implementation developed by the original authors of the EFP
method. The examples used for testing include systems composed of water, ammonia
and benzene molecules in different configurations.
EFP input format
The Q-Chem standard input consists of a number of sections delimited by$keywordand
$end. The mandatory sections are$molecule, which describes the geometry ofab initio
part, and$remsection, which specifies parameters of theab initio computation. Other
sections are specific to different Q-Chem modules.
The EFP code extends the Q-Chem input with 2 new sections:$efp paramsand
$efp fragments, which describe the parameters of the EFP fragment types and define
the molecular system by providing fragments’ positions and orientations, respectively.
In the EFP input sections, all coordinates are in Angstroms, unless Q-Chem’s
INPUT BOHRkeyword is set toTRUE, in which case all the coordinates are in Bohr,
all angles are in radians and other values are in atomic units. All keywords used in the
EFP input section are case insensitive.
Fig. 1.15 presentswhenhe format for the$efp paramssection. It consists of the
parameters’ definitions for one or more fragment types. A definition starts by providing
the name of the fragment type, listing all atoms and their coordinates. It is followed
by the list of all the multipole expansion points (mult). Each multipole expansion point
is defined by its coordinate, multipoles and a damping parameter (DAMP). Multipoles
with different angular momentum are specified on the separate lines in an arbitrary order.
In thier definitionC is charge,D , Q andO are dipole, quadrupole and octupole
components, respectively. All multipoles are optional, but when provided, then all their
33
components should be specified in the order demonstrated in Fig. 1.15. The damping
parameter is optional and should be preceeded by thecdampkeyword.
Next in the fragment definition are the polarization points (pol). They are specified
by the coordinates and components of the polarization tensor P.
The dispersion points (disp) are described similarly by the coordinates and the traces
of the dynamic polarizability tensors Cat 12 different frequencies.
The last subsection describes the parameters required for the exchange-repulsion:
er basis specifies the basis used, erfock provides the Fock matrix, erlmos specifies the
LMO centroids and erwavefunction defines each LMO in terms of the atomic orbitals.
The subsections for different EFP components are optional and can appear in an arbi-
trary order.
Fig. 1.16 presents the format for the$efp fragmentssection. Every effective frag-
ment in the molecular system is defined by the fragment type, followed by its Cartesian
coordinates (X, Y, Z) specifying the translation relative to theab initio system origin
and Euler angles specifying the fragment orientation (α, β, γ). If the orientation is not
provided, the default values of the angles are set to 0. The fragment type should match
one of the fragments in the$efp paramssection.
The active region is specified in the$moleculesection the same way as anab initio
system is specified in Q-Chem47. In order to define a molecular system composed of
the effective fragments only, the keywordEFP FRAGMENTSONLY should be set to
TRUEin the$remsection of the Q-Chem standard inputb.
A complete example of the Q-Chem standard input for the system composed of 3
water molecules (one as an active region and two others as effective fragments) is given
in Appendix F.
bSince the$moleculesection is mandatory in Q-Chem, even if the active region is disabled it shouldstill contain at least one atom, for instance, He. In the nutshell, assigningEFP FRAGMENTSONLYkeyword toTRUEsimply disables any interaction between theab initio part and the effective fragments.
34
$efp params
fragment <fragment type 1>
ATOM X Y Z
...
mult X Y Z
[cdamp DAMP]
[C]
[DX DY DZ]
[QXX QXY QXZ QYY QYZ QZZ]
[OXXX OXXY OXXZ OXYY OXYZ OXZZ OYYY OYYZ OYZZ OZZZ]
...
pol X Y Z
PXX PXY PXZ PYX PYY PYZ PZX PZY PZZ
...
disp X Y Z
S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12
...
er basis BASIS
er fock
F1 F2 F3 F4 ...
er lmo
X Y Z
...
er wavefunction
1 C1 C2 C3 C4 ...
2 C1 C2 C3 C4 ...
...
fragment <fragment type 2>
...
$end
Figure 1.15: The format for the$efp paramssection.
$efp fragments
<fragment type> X Y Z [ α β γ]
...
$end
Figure 1.16: The format for the$efp fragmentssection.
Computing EFP parameters
The EFP parameters part of the code computes the electrostatics, charge-penetration and
polarization parameters for an isolated molecule. The parameters need to be computed35
only once for every fragment type and then can be reused in all subsequent EFP energy
computations.
Since the EFP model has been derived from theab initio theory, the first step in
computing the EFP parameters for a molecule is to obtain its electronic density. This
can be achieved by means of anyab initio method available in Q-Chem that supports
explicit electronic density (for instance: HF, DFT, CCSD).
From the electronic density, the electrostatic parameters (multipoles) are computed
using the distributed multipole analysis procedure as described in section 1.2. The mul-
tipole expansion is computed up to octupoles. The points for the expansion are the
nuclei and the bond midpoints.
The screening parameters accounting for the charge-penetration are computed by
the fitting procedure, which minimizes the difference between the exact electrostatic
potential and potential from the multipoles expansion points multiplied by the screening
function, as detailed in the section 1.2. The optimization of screening parameters is
based on the Newton-Raphson algorithm63.
The polarization parameters are computed using the finite difference method based
on 4ab initio computations: one with no external field and 3 with a small (by default
10−5 V/m) electrostatic field along one of the coordinate axis, as described in section
1.3. In each computation, the localized molecular orbitals are obtained with the Boys
localization procedure57,58.
The computation of the dispersion and exchange-repulsion parameters have not yet
been implemented in Q-Chem. They can be obtained from the quantum chemistry pack-
age GAMESS34, using the code developed by the original authors of the EFP method.
Using itsmakefpjob type, all the EFP parameters can be produced in the GAMESS for-
mat. An example of the GAMESS input file generating the EFP parameters for benzene
and the output produced are shown in Appendix A. To facilitate the conversion from
36
the GAMESS format to the Q-Chem format, a converter script (Appendix C) has been
developed in Python64. It takes a file with the GAMESS EFP parameters as a command
line argument and prints out the parameters in the Q-Chem format. The conversions
performed internally by the script are the syntax conversion, unit conversion (from Bohr
to A) and redundancy elimination (replacing a dynamic polarizability tensor with its
trace).
The Q-Chem keywords affecting the EFP parameters computation are provided in
Appendix G.
1.7 EFP application: Beyond benzene dimer
Aromatic π − π interactions have the major influence on the structure and properties
of many biologically important molecular systems65. They govern nucleotide stacking
patterns in DNA and RNA48, control the ternary structures of proteins66, play an impor-
tant role in the intercalation of drugs with DNA67, affect the properties of the techno-
logically important polymers aramids68, impact the crystal structure of many aromatic
compounds69, and stabilize host—guest complexes70,71.
Studying theπ−π interactions is a major challenge for both experiment and theory72.
The experimental investigations are difficult due to the complex environments in which
these interactions occur (proteins, DNA). The weakness of individualπ−π interactions
and the flatness of the PES along the coordinates defining the relative position of the
aromatic rings make it difficult to compute even for small model systems in the gas
phase73–75.
The theoretical studies ofπ−π interactions, as of any non-covalent interactions, are
challenging due to the weakness of these interactions relative to typical covalent interac-
tions. Their proper treatment requires high accuracyab initio methods [e.g. CCSD(T)]
37
and large basis sets with diffuse functions [e.g. aug-cc-pVTZ]76, which both contribute
to the high computational complexity. At present, high accuracyab initio computations
are feasible only for systems with a fewπ − π interacting aromatic rings.
However, in nature several interactingπ−π moieties are usually present, e.g., many
nucleotides are responsible for the secondary secondary structure of DNA48. Therefore,
it is important to understand whether multipleπ−π interactions can be treated as a sum
of the pair-wise interactions between neighboring aromatic rings or there are significant
non-additive many-body interaction components.
Theoretical studies of multipleπ−π interactions have been so far limited due to the
computational complexity reasons discussed above77–80 . An interesting approach was
taken in the work by Tauer and Sherrill78 who studied benzene trimers and tetramers
using MP2 with a specially selected basis set. MP2 is known to overestimate the binding
energies. For instance, MP2 with a large basis set (aug-cc-pVQZ) overestimates the
dissociation energy of the benezene dimer by 70%76. In contrast, small basis sets are
known to underestimate binding energies. Thus, MP2 with a small basis set may benifit
from the cancelation of errors. The authors selected a basis (cc-pVDZ+78) such that the
errors from MP2 and the basis cancel out each other. Binding energies of the different
benzene dimer configurations at the MP2/cc-pVDZ+ level of theory are within 15%
from the best known theoretical values [CCSD(T)/CBS]. This level of theory has been
used to investigate the nature of binding in benzene trimers and tetramers. The goal
was to understand the importance of the total nonadditive and the individual many-
body interaction energies. The authors concluded that the binding energy of trimers
and tetramers can be described as a simple nearest neighbor sum with 10% accuracy,
and that non-additive (many-body and long range two-body) interactions will become
increasingly important for larger clusters.
38
An important question for theπ − π stacked systems is how the share of the non-
additive energy component grows with the increase in the number of aromatic rings in
the cluster: whether it becomes significant in large clusters or does it stay similar to
that of benzene trimers. A related question is if the binding energy of multipleπ − π
interactions can be well approximated by the sum of the pair-wise nearest neighbor
interaction energies. To answer these fundamental questions, one needs to study a series
of the increasingly large aromatic clusters and observe the asymptotic behavior of the
non-additive energy component. Unfortunately, aromatic clusters with more than a few
benzene rings are beyond the reach of high levelab initio theories. Even using the
proposed in Ref. 78 MP2/cc-pVDZ+ method, computing the energy of the cluster with
a few tens of the benzene rings is presently unfeasible.
The approach to the multipleπ − π interactions problem taken in this work is to
employ the effective fragment potential (EFP) — a semi-empirical method designed
specifically to account for the inter-molecular interactions. In recent works27,28,81, EFP
has been used in the extensive studies of benzene dimers. EFP was found to be in
excellent agreement with high-levelab initio theories. For the three conformers of the
benzene dimer shown in Fig. 1.17, the maximum binding energy was within 0.4 kcal/-
mol and the intermolecular separation was within 0.2A from the values obtained with
CCSD(T)/aug-cc-pVQZ. Additionally, the authors found that the EFP energy profiles
(Fig 10.2 and 10.3 in Ref. 28) for the individual types of interactions (electrostatics,
polarization, dispersion and exchange-repulsion) along the intermolecular separation
coordinates were in excellent agreement with the symmetry-adapted perturbation theory
(SAPT). Thus, not only the total binding energies of the benzene dimers are reproduced
well by EFP, but also the individual types of the interactions are properly accounted for.
39
`
S T PD
RR
R1
R2
Figure 1.17: Sandwitch (S), t-shaped (T) and parallel-displaced (PD) structures of thebenzene dimer.
The discussion above indicates that EFP is an excellent candidate for studying the
systems withπ−π interactions. This work uses our new implementation of EFP to inves-
tigate multipleπ − π interactions. Firstly, the EFP binding energies and geometries for
the sandwich (S), t-shaped (T) and parallel-displaced (PD) structures are compared with
those obtained at differentab initio levels of theory, including the best known to date
theoretical estimates [CCST(T)/CBS]. Then, eight different benzene trimers are studied
in order to understand how well EFP accounts for the non-additive energy contributions.
Finally, the asymptotic behavior of multipleπ− π interactions is investigated in 3 types
of benzene oligomers.
Benzene dimers
The parameters for benzene used throughout this work are those recommended in
Ref. 28. The geometry of the monomer [d(C-C)=1.3942, d(H-H)=1.0823] computed
at the MP2/aug-cc-pVTZ level of theory is from Ref. 82. The effective fragment
potential for the benzene monomer has been generated with GAMESS34 at the HF/6-
311G++(3df,2p) level of theory. As explained in Ref. 26, a basis set of such high qual-
ity is necessary in order to generate accurate dispersion parameters. The input file used
40
for computing EFP parameters is provided in Appendix A. The parameters have been
converted to the Q-Chem format using the script provided in Appendix C.
The benzene dimer structures shown in Fig. 1.17 where constructed analogously to
Ref. 82. The geometries and energies (table 1.1) of 3 benzene dimers corresponding
to the maximum binding at the EFP level of theory were found by performing scans
along the degrees of the freedom shown in Fig. 1.17 with a resolution of 0.1A. At these
geometries, we verified that our implementation computes the total binding energies
and individual energy components correctly by comparing with an analogous GAMESS
computation. The geometries and maximum binding energies we found are only slightly
different (0.1A and 0.15 kcal/mol, respectively) from those obtained in Ref. 28. The
discrepancy is due to the different techniques of the electrostatics screening.c
Table 1.1: The geometric parameters and energies of the benzene dimers
S T PD
Geometric parameters,A R = 4.0 R = 5.2 R1 = 3.9 , R2 = 1.2
Energy, kcal/mol -1.96 -2.38 -2.18
Table 1.2 compares the energies of the 3 benzene dimer structures at different levels
of theory. The best known theoretical estimates are the CCSD(T)/CBS values, which
will serve as a benchmark for our comparison. We can see that CCSD(T)/aug-cc-pVQZ
gives the results within 0.15 kcal/mol from the CBS result. Thus, aug-cc-pVQZ can be
considered sufficient. However, the MP2 results with this basis set are significantly off
(error up to 70%), which is attributed to the overestimation of electron correlation by
MP2 in the benzene dimer83. The method (MP2/aug-cc-pVDZ+) proposed in Ref. 78
cIn Ref. 28 in addition to the screening of charges, which our implementation also performs, the higherorder multipoles (dipoles, quadrupoles, octupoles) were also screened. However, as it was demonstratedin Ref. 27, the accuracy of charge-charge only and charge-charge plus high-order electrostatic screeningsis similar. Therefore, we omitted screening of the higher order multipoles in our implementation.
41
is within 0.4 kcal/mol due to the cancellation of errors. Our EFP results are within 0.6
kcal/mol and thus are of accuracy comparable to MP2/aug-cc-pVDZ+.
Table 1.2: Dissociation energies of the benzene dimers at different levels of theory.
Methoda S T PD
MP2/aug-cc-pVDZ* -2.83 -3.00 -4.12
MP2/aug-cc-pVTZ -3.25 -3.44 -4.65
MP2/aug-cc-pVQZ* -3.35 -3.48 -4.73
MP2/aug-cc-pVQZ -3.37 -3.54 -4.79
MP2/cc-pVDZ+ -1.87 -2.35 -2.84
CCSD(T)/cc-pVDZ* -1.33 -2.24 -2.22
Est. CCSD(T)/aug-cc-pVQZ* -1.70 -2.61 -2.63
Est. CCSD(T)/CBS -1.81 -2.74 -2.78
EFP -1.96 -2.38 -2.18
a The results for theab initio methods are quoted from Ref. 76
Benzene trimers
Using the geometric parameters obtained for dimers (table 1.1), linear benzene trimers
shown in Fig. 1.18 have been constructed. The cyclic structure was obtained by mini-
mizing the binding energy along the intermolecular separation and rotation in the plane
connecting the molecular centers. The minimum corresponds to the molecular sepa-
ratioin of 5.3A, tilted 15 from the perpendicular. The geometries of the trimers are
provided in Appendix B. These trimer structures are analogous to the trimers studied in
Ref. 78 by MP2.
Binding energies for the trimers along with their breakdown into the components are
shown in tables 1.3 and 1.4.E2 is total two-body interaction energy:
E2 = E2(12) + E2(13) + E2(23), (1.39)
42
Table 1.3: Total and many-body interaction energies of the benzene trimers (first group).
S PD T1 T2
EFP
E2(12) -1.96 -2.18 -2.38 -2.40
E2(13) 0.01 0.01 0.02 0.00
E2(23) -1.96 -2.18 -2.40 -2.38
E2 -3.91 -4.35 -4.77 -4.79
E3 0.02 -0.02 0.13 0.06
Etotal -3.89 -4.38 -4.64 -4.73
Etotal(dimer) -3.92 -4.36 -4.76 -4.76
Enon−add 0.03 -0.02 0.12 0.03
∆ 0.77% -0.37% 2.66% 0.57%
MP2/cc-pVDZ+
E3 0.03 0.00 0.08 0.06
Etotal -3.83 -5.88 -4.64 -4.73
Etotal(dimer) -3.74 -5.68 -4.70 -4.70
Enon−add -0.09 -0.20 0.06 -0.03
ECPnon−add 0.03 -0.06 0.1 0.03
∆ 0.78% -1.02% 2.16% 0.63 %
CCSD(T)/cc-pVDZ+
E3 0.04 0.01 0.07
Etotal -0.90 -1.84 -3.14
Etotal(dimer) -0.86 -1.72 -3.20
Enon−add -0.04 -0.12 0.06
ECPnon−add 0.06 0.00 0.10
∆ 6.67% 0.00% 3.18%
E2(NM) - two body interaction between benzenesN and M , E2 - total two bodyinteraction,E3 - three-body interaction,Etotal - total energy of trimer,Etotal(dimer) -total energy of trimer estimated form dimer,Enon−add - non-additive energy component,ECP
non−add - non-additive energy component with counterpoise correction,∆ - share ofthe non-additive energy.
43
Table 1.4: Total and many-body interaction energies of the benzene trimers (secondgroup).
C PD-S S-T PD-T
EFP
E2(12) -2.19 -1.96 -1.96 -2.38
E2(13) -2.19 0.00 -0.12 -0.11
E2(23) -2.19 -2.18 -2.38 -2.18
E2 -6.57 -4.14 -4.46 -4.67
E3 -0.24 0.01 -0.07 -0.06
Etotal -6.81 -4.12 -4.52 -4.73
Etotal(dimer) -6.57 -4.14 -4.34 -4.56
Enon−add -0.25 0.02 -0.18 -0.17
∆ -3.63% 0.43% -4.05% -3.66%
MP2/cc-pVDZ+
E3 -0.33 0.02 -0.03 0.00
Etotal -7.88 -4.86 -4.49 -5.42
Etotal(dimer) -7.32 -4.71 -4.22 -5.19
Enon−add -0.56 -0.15 -0.27 -0.23
ECPnon−add -0.32 -0.01 -0.15 -0.13
∆ -4.06% -0.21% -3.34% -2.40%
CCSD(T)/cc-pVDZ+
E3 -0.25
Etotal -5.09
Etotal(dimer) -4.62
Enon−add -0.47
ECPnon−add -0.26
∆ -5.11%
E2(NM) - two body interaction between benzenesN and M , E2 - total two bodyinteraction,E3 - three-body interaction,Etotal - total energy of trimer,Etotal(dimer) -total energy of trimer estimated form dimer,Enon−add - non-additive energy component,ECP
non−add - non-additive energy component with counterpoise correction,∆ - share ofthe non-additive energy.
44
S T1 T2 PD
C PD-T PD-S S-T
Figure 1.18: The structures of eight benzene trimers.
whereE2(NM) is binding energy for the dimers, created by removing one benzene from
the trimer and computed in the full basis set of the trimer.E3 is three-body interaction
energy computed as:
E3 = Etotal − E2 (1.40)
Etotal(dimer) is total binding energy estimated as the sum of nearest neighbor pair-wise
binding energies.Enon−add andECPnon−add are the non-additive energy components with-
out and with the counterpoise correction:
Enon−add = Etotal − Etotal(dimer)
ECPnon−add = Etotal − ECP
total(dimer)
(1.41)
45
∆ is the share of the non-additive interaction energy which EFP computes as:
∆ =Enon−add
Etotal
(1.42)
andab initio methods as:
∆ =ECP
non−add
Etotal
(1.43)
To evaluate how well EFP describes multipleπ − π interactions, it is necessary to
conduct a comparison withab initio methods. The quality of the description of the
two-body interactions (between two benzene rings) has been investigated in the exten-
sive studies of the benzene dimers energy profiles in Ref. 28. The conclusion was that
EFP results are in good agreement with the high levelab initio theories [CCSD(T) and
SAPT]. Three-body interactions for EFP, MP2/cc-pVDZ+ and CCSD(T)/cc-pVDZ+ are
compared in tables 1.3 and 1.4. The first important observation is that theab initio
E3 values quoted from Ref. 78 were computed using the counterpoise correction to
account for the basis set superposition error, which results from the basis set incom-
pleteness. Three-body interaction energies were calculated by subtracting from total
binding energy of the trimer energies of the 3 possible benzene dimers computed in the
full set basis of the trimer.
All EFP many-body interactions are due to the polarization component. Thus, itsE3
term is three-body polarization energy. From tables 1.3 and 1.4, we can see that three-
body interaction energies computed with EFP have the same order of magnitude as for
theab initio methods. For most trimers studied here, theE3 correction is rather small
and miscellaneous errors may have a significant impact on its value. The exception is the
cyclic benzene trimer (see Fig. 1.18), which is a true three-body system demonstrating
substantial three-body interactions at all the considered levels of theory. For this trimer
E3 computed with EFP is within 0.01 kcal/mol from the CCSD(T) value. The fact that
46
these values almost coincide might be accidental, but their closeness implies a certain
level of accuracy in the treatment of three-bodyπ− π interactions by EFP. On the basis
of the above comparison, we conclude that EFP provides a good description of three-
body interactions in benzene trimers.
Another interesting conclusion can be drawn from the comparison of total non addi-
tive interaction energies (Enon−add). Firstly, Enon−add for EFP are greater than for
CCSD(T) and MP2 for all eight considered trimers. However, we concluded that this
effect is mostly due to the basis set superposition error (BSSE).Enon−add is computed
as a difference between trimer binding energy and energies of two dimers (which were
derived from the trimer by removing one of the edge benzenes). Thus, the basis sets
used for the trimer and the dimers energy computations are different. Since trimer has
a larger basis set, its nearest neighborπ − π interactions are stronger than in the corre-
sponding dimers. That is, the additional basis functions contribute to the wavefunction
relaxation, leading to binding energy lowering. In fact, table 1 of Ref. 78 shows the
energies of the dimers computed in the full basis set of the trimer. These energies are
0.05-0.10 kcal/mol lower than for the dimers in its own basis set. Thus, as much as 0.3
kcal/mol for the cyclic trimer and 0.2 kcal/mol for other trimers can be contributed to
the lowering ofEnon−add due to the basis set relaxation effect. Since EFP does not suffer
from the basis set relaxation, itsEnon−add for benzene trimers are systematically higher.
Enon−add computed byab initio methods can be corrected by using the binding ener-
gies of the dimers computed in the full basis set of the trimer, similarly to the counter-
poise correction often employed to rectify BSSE. Since these values are readily available
in table 1 of Ref. 78, we were able to compute corrected non-additive interaction energy
ECPnon−add, which is also provided in tables 1.3 and 1.4. EFP non-additive interaction
energies are within 0.03 kcal/mol fromECPnon−add for CCSD(T) and within 0.04 kcal/-
mol from MP2 values (except for the cyclic trimer). The share of the total non-additive
47
energy correction∆ does not exceed 4% for EFP and MP2. This share is larger for
CCSD(T) because the method significantly underestimates the total binding energy. It
also may attributed to the inclusion of higher-order three-body interaction energy terms
(e.g., three-body dispersion) in CCSD(T).
Based on the above comparisons of three-body and total non-additive interaction
energies, we conclude that EFP accurately describes the multipleπ − π interactions in
the considered benzene trimers.
Large benzene oligomers
The last part of our investigation is a study of the asymptotic behavior of the non-additive
energy contribution. Using the intermolecular separation parameters for dimers (table
1.1), we constructed the linear oligomers with 10, 20, 30 and 40 benzene molecules.
The oligomers S-N, T-N and PD-N, where N is the number of monomers, have the same
structural motives as trimers S, T1 and PD shown in Fig. 1.18. The geometries of the
oligomers are provided in Appendix B. The total binding energies (Etotal), the nearest
neighbor estimates (Etotal(dimer)), the non-additive energy (Enon−add) and its share (∆)
in the total energy are shown in table 1.5. The nearest neighbor estimate for the oligomer
with N benzene rings has been computed as:
Etotal(dimer) = (N − 1)Edimer (1.44)
From table 1.5, we can see that for all the oligomersEnon−add is positive and thus
destabilizing in nature, as it was correctly predicted in Ref. 78. It is also in good
agreement with the observation that nonadditive energy is positive in the bulk benzene
made in Ref. 84. Non-additive energy grows with the size of the system because of the
48
Table 1.5: Binding energies of benzene oligomers computed with EFP.
Etotal Etotal(dimer) Enon−add ∆
S-10 -17.14 -17.64 0.50 2.94%
S-20 -36.01 -37.24 1.23 3.41%
S-30 -54.89 -56.84 1.95 3.56%
S-40 -73.76 -76.44 2.68 3.63%
T-10 -20.89 -21.42 0.53 2.53%
T-20 -44.05 -45.22 1.17 2.65%
T-30 -67.21 -69.02 1.81 2.69%
T-40 -90.37 -92.82 2.45 2.71%
PD-10 -19.54 -19.62 0.08 0.41%
PD-20 -41.16 -41.42 0.26 0.63%
PD-30 -62.78 -63.22 0.44 0.70%
PD-40 -84.40 -85.02 0.62 0.73%
a S-N, T-N and PD-N are linear benzene oligomers with N benzene monomers, con-structed using sandwich, t-shaped and parallel-displaced structural motives, respec-tively.
large number of many-body and long-range two-body interactions possible in the larger
oligomers.
The dependence of the share of non-binding energy∆ on the system size is shown
in Fig. 1.19. For all 3 types of the oligomers,∆ asymptotically approaches a constant
value and does not exceed 4% of the total binding energy.
Based on the investigation of benzene trimers and higher-order oligomers at the EFP
level of theory, we conclude that the share of total non-additive energy is about 4%
for the systems considered. This result implies that total interaction energy in clustered
π−π systems can be estimated reasonably well as the sum of two-body nearest neighbor
interactions. However, it is important to note that neither EFP applied in our work
nor MP2 used in the studies of trimers in Ref. 78 account for many-body dispersion.
As demonstrated in the recent study79, many-body dispersion terms contribute about
49
1 0 2 0 3 0 4 00 . 5
1 . 0
1 . 5
2 . 0
2 . 5
3 . 0
3 . 5
Share
of th
e non
-addit
ive en
ergy,
%
T h e n u m e r o f b e n z e n e m o n o m e r s
S P D T
Figure 1.19: Dependence of the non-additive energy share on the length of benzeneoligomers.
50% to total three-body interaction energy of the cyclic benzene trimer. Additionally,
we expect that the share of the non-additive energy contributions will increase in the
three-dimensional clusters and bulk benzene due to the large number of many-body
interactions, which can occur between molecules located within a short distance from
each other.
1.8 Conclusions
This chapter presented a review of the theoretical approaches to modeling the envi-
ronmental effects and a detailed description of the EFP method. The details of our
EFP implementation in the Q-Chem package were presented. The developed code was
50
applied to investigate multipleπ − π interactions in benzene oligomers focusing on the
role of many-body and total non-additive energies in these systems.
We analyzed previously reported78 ab initio results for benzene trimers and con-
cluded that they underestimate non-additive energy due to the basis set superposition
error. Based on the data presented in Ref. 78, we were able to recomputeab initio
non-additive energy with the counterpoise correction. Using our EFP code, we com-
puted three-body and total non-additive energies for benzene trimers. These energies
for all eight timers included in our study appeared to be in excellent agreement with the
counterpoise-correctedab initio values (within 0.04 kcal/mol). The magnitude of non-
additive interaction energy was within 4% of total binding energy for all the considered
benzene clusters. Based on the detailed comparison of the binding energy components
(presented in tables 1.3 and 1.4), we concluded that EFP describes the intermolecular
π − π interactions with the accuracy comparable to that ofab initio methods.
In the final part, our EFP code was applied to study linear benzene oligomers. Based
on the results (presented in table 1.5), we concluded that the share of non-additive inter-
action energy in linear benzene clusters asymptotically approaches a constant value with
the increase in cluster size, and does not exceed 4% of total binding energy. Thus, it is
approximately of the same magnitude as for the benzene trimers. This result implies that
there are no significant non-local effects in the binding mechanism of the linear benzene
clusters.
Reference list
[1] Raghavachari, K. ; Trucks, G.W. ; Pople, J.A. ; Head-Gordon, M.Chem. Phys.Lett.1989, 157, 479.
51
[2] Bowler, D.R. ; Miyazaki, T. ; Gillan, M.J.J. Phys.: Condens. Matter2002, 14,2781.
[3] Schwegler, E. ; Challacombe, M.J. Chem. Phys.1996, 105, 2726.
[4] White, C.A. ; Johnson, B.G. ; Gill, P.M.W. ; Head-Gordon, M.Chem. Phys. Lett.1996, 253, 268.
[5] Schutz, M. ; Hetzer, G. ; Werner, H.J. Chem. Phys.1999, 111, 5691.
[6] Scuseria, G.E. ; Ayala, P.Y.J. Chem. Phys.1999, 111, 8330.
[7] Gogonea, V. ; Westerhoff, L.M. ; K.M. Merz, Jr.J. Chem. Phys.2000, 113, 5604.
[8] Kitaura, K. ; Ikeo, E. ; Asada, T. ; Nakano, T. ; Uebayasi, M.Chem. Phys. Lett.1999, 313, 701 .
[9] Lippert, G. ; Hutter, J. ; Parrinello, M.Mol. Phys.1997, 92, 477.
[10] Allunger, N.L. ; Yuh, Y.H. ; Lii, J.H.J. Am. Chem. Soc.1989, 111, 8551.
[11] Casewit, C.J. ; Colwell, K.S. ; Rappe, A.K.J. Am. Chem. Soc.1992, 114, 10035.
[12] Mayo, S.L. ; Olafson, B.D. ; Goddard, W.A.J. Phys. Chem.1990, 94, 8897.
[13] Rizzo, R.C. ; Jorgensen, W.L.J. Am. Chem. Soc.1999, 121, 4827.
[14] Firman, T.K. ; Landis, C.R.J. Am. Chem. Soc.2001, 123, 11728.
[15] Jorgensen, W.L. ; McDonald, N.A.J. Molec. Struct. (Theochem)1998, 424, 145 .
[16] MacKerell, A.D. ; Bashford, D. ; Bellott; Dunbrack, R.L. ; Evanseck, J.D. ; Field,M.J. ; Fischer, S. ; Gao, J. ; Guo, H. ; Ha, S. ; Joseph-McCarthy, D. ; Kuchnir, L.; Kuczera, K. ; Lau, F.T.K. ; Mattos, C. ; Michnick, S. ; Ngo, T. ; Nguyen, D.T. ;Prodhom, B. ; Reiher, W.E. ; Roux, B. ; Schlenkrich, M. ; Smith, J.C. ; Stote, R.; Straub, J. ; Watanabe, M. ; Wiorkiewicz-Kuczera, J. ; Yin, D. ; Karplus, M.J.Phys. Chem. B1998, 102, 3586.
[17] Hill, J.R. ; Sauer, J.J. Phys. Chem.1995, 99, 9536.
[18] Lin, H. ; Truhlar, D.G.Theor. Chem. Acc.2007, 117, 185.
[19] Gao, J.L. ; Truhlar, D.G.Annu. Rev. Phys. Chem.2002, 53, 467.
[20] Sierka, M. ; Sauer, J.J. Chem. Phys.2000, 112, 6983.
[21] Monard, G. ; Merz, K.M.Acc. Chem. Res.1999, 32, 904.
52
[22] Day, P.N. ; Jensen, J.H. ; Gordon, M.S. ; Webb, S.P. ; Stevens, W.J. ; M.Krauss;D.Garmer; H.Basch; D.CohenJ. Chem. Phys.1996, 105, 1968.
[23] Jensen, J.H. ; Gordon, M.S.J. Chem. Phys.1998, 108, 4772.
[24] Freitag, M.A. ; Gordon, M.S. ; Jensen, J.H. ; Stevens, W.J.J. Chem. Phys.2000,112, 7300.
[25] Gordon, M.S. ; Freitag, M.A. ; Bandyopadhyay, P. ; Jensen, J.H. ; Kairys, V. ;Stevens, W.J.J. Phys. Chem. A2001, 105, 293.
[26] Adamovix, I. ; Gordon, M.S.Mol. Phys.2005, 103, 379.
[27] Slipchenko, L.V. ; Gordon, M.S.J. Comput. Chem.2007, 28, 276.
[28] Gordon, M.S. ; Slipchenko, L. ; Li, H. ; Jensen, J.H. InAnnual Reports in Com-putational Chemistry; Spellmeyer, D.C. , Wheeler, R. , Eds., Vol. 3 ofAnnualReports in Computational Chemistry;Elsevier, 2007; pages 177 – 193.
[29] Slipchenko, L.V. ; Gordon, M.S.Mol. Phys.2009.
[30] Li, H. ; Netzloff, H.M. ; Gordon, M.S.J. Chem. Phys.2006, 125.
[31] Li, H. ; Gordon, M.S. ; Jensen, J.H.J. Chem. Phys.2006, 124.
[32] Li, H. ; Gordon, M.S.J. Chem. Phys.2007, 126.
[33] Li, H. ; Gordon, M.S.Theor. Chem. Acc.2006, 115, 385.
[34] Schmidt, M.W. ; Baldridge, K.K. ; J.A. Boatz, S.T. Elbert ; Gordon, M.S. ;J.H. Jensen, S. Koseki ; Mastunaga, N. ; Nguyen, K.A. ; S. Su, T.L. Windus ;Dupuis, M. ; Montgomery, J.A.J. Comput. Chem.1993, 14, 1347.
[35] Adamovic, I. ; Gordon, M.S.J. Phys. Chem. A2005, 109, 1629.
[36] Balawender, R. ; Safi, B. ; Geerlings, P.J. Phys. Chem. A2001, 105, 6703.
[37] Chandrakumar, K.R.S. ; Ghanty, T.K. ; Ghosh, S.K. ; Mukherjee, T.J. Molec.Struct. (Theochem)2007, 807, 93.
[38] Day, P.N. ; Pachter, R.J. Chem. Phys.1997, 107, 2990.
[39] Merrill, G.N. ; Webb, S.P.J. Phys. Chem. A2004, 108, 833.
[40] Yoo, S. ; Zahariev, F. ; Sok, S. ; Gordon, M.S.J. Chem. Phys.2008, 129, 8.
[41] Freitag, M.A. ; Gordon, M.S.Abstracts of Papers of the American Chemical Soci-ety1999, 218, U511; 1.
53
[42] Merrill, G.N. ; Gordon, M.S.J. Phys. Chem. A1998, 102, 2650.
[43] Merrill, G.N. ; Webb, S.P.J. Phys. Chem. A2003, 107, 7852.
[44] Merrill, G.N. ; Webb, S.P. ; Bivin, D.B.J. Phys. Chem. A2003, 107, 386.
[45] Netzloff, H.M. ; Gordon, M.S.J. Chem. Phys.2004, 121, 2711.
[46] Day, P.N. ; Pachter, R.and Gordon, M.S. ; Merrill, G.N.J. Chem. Phys.2000, 112,2063.
[47] Shao, Y. ; Molnar, L.F. ; Jung, Y. ; Kussmann, J. ; Ochsenfeld, C. ; Brown, S. ;Gilbert, A.T.B. ; Slipchenko, L.V. ; Levchenko, S.V. ; O’Neil, D. P. ; Distasio Jr.,R.A. ; Lochan, R.C. ; Wang, T. ; Beran, G.J.O. ; Besley, N.A. ; Herbert, J.M. ; Lin,C.Y. ; Van Voorhis, T. ; Chien, S.H. ; Sodt, A. ; Steele, R.P. ; Rassolov, V. A. ;Maslen, P. ; Korambath, P.P. ; Adamson, R.D. ; Austin, B. ; Baker, J. ; Bird, E.F.C.; Daschel, H. ; Doerksen, R.J. ; Drew, A. ; Dunietz, B.D. ; Dutoi, A.D. ; Furlani,T.R. ; Gwaltney, S.R. ; Heyden, A. ; Hirata, S. ; Hsu, C.-P. ; Kedziora, G.S. ;Khalliulin, R.Z. ; Klunziger, P. ; Lee, A.M. ; Liang, W.Z. ; Lotan, I. ; Nair, N. ;Peters, B. ; Proynov, E.I. ; Pieniazek, P.A. ; Rhee, Y.M. ; Ritchie, J. ; Rosta, E. ;Sherrill, C.D. ; Simmonett, A.C. ; Subotnik, J.E. ; Woodcock III, H.L. ; Zhang, W.; Bell, A.T. ; Chakraborty, A.K. ; Chipman, D.M. ; Keil, F.J. ; Warshel, A. ; Herhe,W.J. ; Schaefer III, H.F. ; Kong, J. ; Krylov, A.I. ; Gill, P.M.W. ; Head-Gordon, M.Phys. Chem. Chem. Phys.2006, 8, 3172.
[48] Saenger, W.Principles of Nucleic Acid Structure;Springer-Verlag: New York,1984.
[49] Buckingham, A.D.Q. Rev. Chem. Soc.1959, 13, 183.
[50] Stone, A.J.Chem. Phys. Lett.1981, 83, 233 .
[51] Mulliken, R.S.J. Chem. Phys.1955, 23, 1833.
[52] Stone, A.J. ; Alderton, M.Mol. Phys.1985, 56, 1047.
[53] Stone, A.J. InThe Theory of Intermolecular Forces;Oxford University Press,1997; pages 50 – 62.
[54] Stone, A.J.J. Chem. Theory Comput.2005, 1, 1128.
[55] Rabinowitz, J.R. ; Namboodiri, K. ; Weinstein, H.Int. J. Quant. Chem.1986, 29,1697.
[56] Wolfsberg, M. ; Helmholz, L.J. Chem. Phys.1952, 20, 837.
[57] Boys, S.F.Rev. Mod. Phys.1960, 32, 296.
54
[58] Boys, S.F. InQuantum Science of Atoms, Molecules, and Solids;Academic Press,1966; pages 253–262.
[59] Sanz-Garcia, A. ; Wheatley, R.J.ChemPhysChem2003, 5, 801.
[60] Tang, K.T. ; Toennies, J.P.J. Chem. Phys.1984, 80, 3726.
[61] Jensen, J.H. ; Gordon, M.S.Mol. Phys.1996, 89, 1313.
[62] ISO/IEC ISO/IEC 14882:2003: Programming languages: C++ Technical report,International Organization for Standardization, Geneva, Switzerland.,2003.
[63] Press, W.H. ; Teukolsky, S.A. ; Vetterling, W.T. ; Flannery, B.P.Numerical Recipes3rd Edition: The Art of Scientific Computing;Cambridge University Press: NewYork, NY, USA, 2007.
[64] Rossum, G.V.The Python Language Reference Manual;Network Theory Ltd.,2003.
[65] Hunter, C.A. ; Sanders, J.K.M.J. Am. Chem. Soc.1990, 112, 5525.
[66] Burley, S.K. ; Petsko, G.A.Science1985, 229, 23.
[67] Brana, M.F. ; Cacho, M. ; Gradillas, A. ; Pascual-Teresa, B. ; Ramos, A.Curr.Pharm. Des.2001, 7, 1745.
[68] Gazit, E.Chem. Soc. Rev.2007, 36, 1263.
[69] Desiraju, G.R. ; Gavezzotti, A.JCS Chem. Comm.1989, page 621.
[70] Askew, B. ; Ballester, P. ; Buhr, C. ; Jeong, K.S. ; Jones, S. ; Parris, K. ; Williams,K. ; Rebek, J.J. Am. Chem. Soc.1989, 111, 1082.
[71] Smithrud, D.B. ; Diederich, F.J. Am. Chem. Soc.1990, 112, 339.
[72] Muller-Dethlefs, K. ; Hobza, P.Chem. Rev.2000, 100, 143.
[73] Felker, P.M. ; Maxton, P.M. ; Schaeffer, M.W.Chem. Rev.1994, 94, 1787.
[74] Leopold, K.R. ; Fraser, G.T. ; Novick, S.E. ; Klemperer, W.Chem. Rev.1994, 94,1807.
[75] Mueller-Dethlefs, K. ; Dopfer, O. ; Wright, T.G.Chem. Rev.1994, 94, 1845.
[76] Sinnokrot, M.O. ; Sherrill, C.D.J. Phys. Chem. A2006, 110, 10656.
[77] Ye, X. ; Li, Z. ; Wang, W. ; Fan, K. ; Xu, W. ; Hua, Z.Chem. Phys. Lett.2004, 397,56 .
55
[78] Tauer, T.P. ; Sherrill, C.D.J. Phys. Chem. A2005, 109, 10475.
[79] Podeszwa, R. ; Szalewicz, K.J. Chem. Phys.2007, 126, 194101.
[80] Podeszwa, R.J. Phys. Chem. A2008, 112, 8884.
[81] Smith, T. ; Slipchenko, L.V. ; Gordon, M.S.J. Phys. Chem. A2008, 112, 5286.
[82] Sinnokrot, M.O. ; Valeev, E.F. ; Sherrill, C.D.J. Am. Chem. Soc.2002, 124, 10887.
[83] Tsuzuki, S. ; Uchimaru, T. ; Matsumura, K. ; Mikami, M. ; Tanabe, K.Chem.Phys. Lett.2000, 319, 547 .
[84] Podeszwa, R. ; Bukowski, R. ; Szalewicz, K.J. Phys. Chem. A2006, 110, 10345.
56
Chapter 2
Structure, vibrational frequencies,
ionization energies, and photoelectron
spectrum of the para-benzyne radical
anion
2.1 Introduction
Para-benzyne1, an intermediate in Bergman cyclization2, is believed to be a warhead in
antitumor enediyne antibiotics3–5 due to its ability to cause double-strand DNA cleavage
and self-programmed cell death or apoptosis6,7, and extensive studies have been carried
out in the past 20 years to understand the factors that control the reaction8–10. Originally,
it was interest in para-benzyne electronic structure that motivated the development of
procedures to generate the corresponding anion for use in photoelectron spectroscopy
experiments11,12. Recent studies suggested that anionic cycloaromatization reactions
are also feasible13,14 including the anionic version of the Bergman cyclization15. Larger
sensitivity to substituents15 suggests that the reactions of the anions might be designed
to be more selective than the prototypic non-anionic Bergman cyclization, important for
their applications in cancer-DNA damaging drugs.
Previous theoretical studies of the para-benzyne anion16 predicted strong vibronic
interactions (often referred to as pseudo or second-order Jahn-Teller effect) that produce
57
a lower symmetryC2v structure in addition to theD2h minimum. The energy difference
between the two structures was found to be highly sensitive to the correlation method,
with higher-level models favoringD2h. The analysis of the photoelectron spectrum of
the anion12 supported the assignment of theD2h symmetry, and it was concluded that
the lower symmetry structures are artifacts of incomplete correlation treatment, a well
known phenomenon in electronic structure17. Interestingly, density functional calcula-
tions produced onlyD2h structures, which was regarded as success of DFT methodol-
ogy and the authors concluded that “this level of theory is particularly well-suited for
computational studies of distonic radical anions derived from diradicals”16. The robust
behavior of DFT in cases of symmetry breaking has been noted by other researches as
well18.
The goal of this work is to investigate the performance of coupled-cluster (CC) and
equation-of-motion (EOM) methods in this challenging case of strong vibronic interac-
tions and to assess the quality of theab initio and DFT results by comparing the cal-
culated photoelectron spectrum with the experimental one. We quantify the symmetry
lowering by the selected geometrical parameters, charge distributions, and relative ener-
gies. In addition, the shape of the potential energy surface along the symmetry breaking
coordinate is analyzed. We also present an efficient scheme for calculating ionization
energies (IEs) of the anion in the spirit of isodesmic reactions.
Several recent studies19–21discussed different aspects of symmetry breaking in elec-
tronic structure calculations distinguishing between: (i) purely artefactual spatial and
spin symmetry breaking of approximate wave functions, i.e., Lowdin dilemma22; (ii)
real interactions between closely-lying electronic states that result in lower-symmetry
structures, significant changes in vibrational frequencies (see Fig. 1 from Ref. 21), and
even singularities (first order poles) in force constants (e.g., Fig. 3 from Ref. 20). While
the latter is a real physical phenomenon, it may or may not be accurately described by an
58
approximate method. If vibronic interactions are overestimated, calculations may yield
incorrect lower-symmetry structure, and vice verse.
Formally, the effect of the interactions between electronic states on the shape of
potential energy surface can be described20 by the Herzberg-Teller expansion of the
potential energy of electronic statei:
Vi = V0 +∑
α
< Ψi|∂H
∂Qα
|Ψi > Qα+
1
2
∑α
< Ψi|∂2H
∂2Qα
|Ψi > Q2α−
∑α
∑j 6=i
(< Ψi| ∂H
∂Qα|Ψj >
)2
Ej − Ei
Q2α,
(2.1)
whereQα denotes normal vibrational modes, wave functionsΨk and energiesEk are
adiabatic wave functions and electronic energies atQα = 0. The last term, which is
quadratic in nuclear displacement — hencesecond-order Jahn-Teller, describes vibronic
effects. For the ground state, i.e., whenEj > Ei, it causes softening of the force constant
alongQα, which may ultimately result in a lower symmetry structure, e.g., see Fig. 1
from Ref. 21. The above mentioned poles in force constants originate in the energy
denominator.
The formal analysis of the CC and EOM-CC second derivatives by Stanton20
demonstrated that the quadratic force constant of standard coupled-cluster methods,
e.g., coupled-cluster with singles and doubles (CCSD), necessarily contains unphysi-
cal terms, which may become significant and spoil the CC force constant in the cases
of strongly interacting states. Nevertheless, the standard CC methods are quite reliable
for systems exhibiting pseudo Jahn-Teller interactions, as follows from Refs. 20, 21, as
well as a host of numerical evidence. EOM-CC methods, on the other hand, provide
most satisfactory description of the interacting states20, not at all surprising in view of a
59
multistage nature of EOM. It should be noted that both CC and EOM-CC may exhibit
spurious frequencies if the orbital response poles, i.e., coupled-perturbed Hartree-Fock
(CPHF) poles, are present due to near-instabilities in the reference, a condition closely
related to (i). Perturbative corrections, e.g., as in CCSD(T), do not provide systematic
improvement, and often results in wider instability volcanoes19.
By analyzing the properties of linear response of multi-configurational SCF
(MCSCF) and the relationship between response equations and poles’ structure, Stanton
dispelled a common misconception of relying on multi-reference treatment of systems
with strong vibronic interactions20. Using the same arguments, he also concluded that
DFT describes pseudo Jahn-Teller effects reasonably well. While some numerical evi-
dence18 seems to support this conclusion, other examples21 indicated that DFT (B3LYP)
tend tounderestimatethe magnitude of vibronic interactions. Our results are in favor of
the latter, and we attribute this behavior to self-interaction error.
The structure of this chapter is as follows. The next section describes molecular
orbitals (MOs) and relevant electronic states of the para-benzyne anion, as well as the
nature of vibronic interactions in this system. Section 2.3 outlines computational details.
Section 2.4 discusses the competition between lower and higher symmetry structures at
different levels of theory. Photodetachment spectra are presented in section 2.5. Differ-
ent computational strategies for calculating IEs are discussed in section 2.7. Our final
remarks are given in the last section.
2.2 Molecular orbital picture and the origin of vibronic
interactions in the para-benzyne anion
The frontier MOs of the para-benzyne anion and the diradical are shown in Fig. 2.1. At
D2h, the two orbitals belong to different irreps,b1u andag, and their density is equally
60
distributed between the two radical centers (C1 and C4). At C2v, both orbitals become
a1, and their densities are localized at different carbons. As in the benzyne diradical,
the lowest MO, which hosts two electrons in the anion, is of anti-bonding character with
respect to C1-C4, whereas the singly occupied MO is bonding. In the case of the dirad-
ical, three singlet states and one triplet state are derived from different distributions of
two electrons on these two orbitals: X1Ag, 21Ag, 11B1u, 23B1u. The ground singlet state
belongs to the same irrep as the doubly excited singlet, and both have significant multi-
configurational character. Likewise, distributing three electrons on these two orbitals in
the anion gives rise to two electronic configurations shown in the lower panel of Fig. 2.1.
At D2h, these two determinants are of different symmetry and correspond to two distinct
electronic states, X2Ag and2B1u, separated by the energy gap of about 0.95 eV, as cal-
culated by EOM-CC for electron-attached states (see below). AtC2v, both determinants
are ofA1 symmetry and, therefore, can interact. This interaction increases the energy
separation between the two electronic states that acquire multi-configurational character
and, therefore, “softens” the corresponding normal mode in the lower state. If the inter-
action is sufficiently strong, that is, if the coupling matrix element is large relative to the
energy gap between the interacting states, the lower potential energy surface (PES) may
develop a lower symmetry minimum. In other words, lowering symmetry stabilizes the
lowest electronic state through configurational interaction, and this is the driving force
for symmetry breaking.
This analysis of the wave functions of interacting states complements the lessons
learned from Eq. (2.1) and poles’ structure of quartic force constants20,21. For example,
it shows that two factors are important for accurate description of systems with consid-
erable vibronic interactions: (i) balanced description of possibly multi-configurational
wave function (non-dynamical correlation); and (ii) quantitative accuracy in describing
61
C2v2hD
a1
a1
ag
b1u
C1
C1 C1
C1
C4 C4
C4 C4
X2Ag
A2B1u
X2A1
A2A1
6ag
5b1u
6
5
6ag
5b1u
g
5
11a1
10a1
11
10
11a1
10a1
11
10
Figure 2.1: Frontier MOs and leading electronic configurations of the two lowest elec-tronic states atD2h andC2v structures.
62
excited state ordering (dynamical correlation). Unbalanced treatment may overempha-
size the magnitude of vibronic interaction and, therefore, may lead to so-called arte-
factual symmetry breaking17. While (i) can be achieved by multi-configurational self-
consistent field (MCSCF), (ii) is much more difficult to satisfy. For example, MCSCF
exhibits artificial symmetry breaking in NO3, which is removed when dynamical cor-
relation is included23. Ground-state coupled-cluster methods, e.g., CCSD or CCSD(T),
may fail due to the insufficient description of non-dynamical correlation, as it happens
in NO324,25, although they often tackle such challenging systems successfully26.
The analysis of the MOs and the electronic states of the para-benzyne anion
(Fig. 2.1) suggests a moderate multi-configurational character of the ground state wave
function, and this is confirmed by EOM-EA amplitudes (as calculated at EOM-EA-
CCSD/6-311+G**) For theD2h structure, the EOM coefficient of the leading config-
uration (theAg deteminant from Fig. 2.1 and 2.2) is 0.86. The three next important
configurations have weights of 0.28, 0.26 and 0.07. Likewise, for theC2v structure the
weights of the four leading configurations are 0.87, 0.27, 0.16 and 0.15. The EOM-
SF-CCSD/6-311+G** amplitudes are similar: the weights of the leading configurations
are 0.81 and 0.80 for theD2h andC2v structures, respectevily. This confirms minor
multi-configurational character of the para-benzyne anion. For comparison, the weights
of the four leading configurations of the singlet para-benzyne diradical, which is a truly
multiconfigurational system, are 0.57, 0.38, 0.32 and 0.31 and the two leaing configu-
rations are double excitations with respect to each other. Thus, we expect ground-state
CC methods to be capable of treating C6H−4 reasonably well, unlike heavily multicon-
figurational wavefunctions, as those of the singlet diradicals24.
The EOM methods that simultaneously include both dynamical and non-dynamical
correlation are particularly attractive for the systems with vibronic interactions. More-
over, as formal analysis of Stanton shows, the pole structure of the exact force constant
63
6ag
5b1u
6
5
6ag
5b1u
g
5
6ag
5b1u
g
5
6ag
5b1u
g
5
Reference:
7agg
p-C H 62-4
Quartet
Tripletp-C H 6 4
p-C H 6-4
Singlet
+
Doublet p-C H 6-4
Ψtarget = R IP Ψ
ref
Ψtarget = REA Ψ ref
Ψ target = RSF Ψ ref
Figure 2.2: The two lowest electronic states of C6H−4 can be described by: (i) EOM-IP
with the dianion reference; (ii) EOM-EA with the triplet neutral p-benzyne reference;(iii) EOM-SF with the quartet p-benzyne anion reference.
of Eq. (2.1) is correctly reproduced within EOM-CC formalism. From the configura-
tion interaction point of view, the balanced description of two important determinants
from Fig. 2.1 is easily provided by several EOM-CC methods, i.e., ionization poten-
tial, electron-attachment, and spin-flip EOM-CC (EOM-EA, EOM-IP, and EOM-SF,
respectively), as explained in Fig. 2.2. All three EOM models include dynamical corre-
lation as well. Their reliability depends on how well the corresponding reference state is
described by the single-reference CCSD method and possible (near)-instabilities of the
Hartree-Fock references.
64
2.3 Computational details
Two basis sets were employed in this work. All the geometry optimizations, fre-
quencies’ calculations and most single point calculations were performed with the 6-
311+G**27,28basis. Additional single point calculations of IEs used the aug-cc-pVTZ29
basis.
The choice of reference in our calculations requires additional comments. The
ground-state methods [i.e., MP2, CCSD, CCSD(T)], which employ2Ag reference, were
conducted using pure-symmetry ROHF and UHF references. The< S2 > values for
UHF were 0.869 and 1.226 atD2h andC2v, respectively. The stability analysis revealed
several symmetry-breaking triplet instabilities atD2h. The lowest-energy UHF solution
is heavily spin-contaminated,< S2 >=1.248, and is also of broken symmetry. When
employed in the CCSD calculations, the lower-energy broken-symmetry UHF reference
yielded higher correlated energies, in agreement with our experience (see also foot-
note 17 in Ref. 20). Thus, we consider symmetry-pure UHF references to be more
appropriate in CC calculations, even though this results in cusps in the PES scans along
theD2h→ C2v distortion. Using symmetry-broken reference atD2h yields continuous
curves. In the cases of moderate spin-contamination (as for pure-symmetry UHF refer-
ences), CCSD is orbital insensitive, however, the (T) correction is often more reliable
for ROHF (see, for example, results from Ref. 30). The optimized-orbital CC models
avoid ambiguity of the reference choice for non-variational CCSD, and, in view of UHF
instabilities, we consider optimized orbitals coupled-cluster with doubles (OO-CCD)
reference to be most appropriate (among the ground-state CC methods) for this system.
References for the EOM calculations (see Fig. 2.2) were not well behaved either.
The closed-shell1A1 reference for EOM-IP corresponds to the C6H2−4 dianion, which is
unstable. This caused severe convergence problems and we abandoned EOM-IP treat-
ment. The EOM-EA calculations employed the3B1u state (3A1 atC2v) of the diradical.
65
Unexpectedly, the corresponding UHF reference develops strong spin contamination
at relatively small displacements fromD2h: for example,< S2 >=2.026 and 2.468
andD2h andC2v, respectively. This compromises the reliability of EOM-EA-CCSD at
C2v. The problem can be resolved by performing the EOM-EA calculations using the
OO-CCD triplet reference (EOM-EA-OD). The EOM-SF calculations employ the4B1u
reference (4A1 atC2v), which can be derived from the3B1u configuration by placing an
additional electron on aσ-like ag orbital. Despite the similarity to the parent triplet, the
quartet reference has been found to be well behaved throughout the whole range of the
D2h→C2v distortions, e.g.,< S2 >=3.776 and 3.779 andD2h andC2v, respectively.
Optimized geometries were obtained at the UHF, ROHF, UHF-CCSD31, UHF-
CCSD(T)32, OO-CCD31,33, EOM-EA-CCSD34, EOM-SF-CCSD35,36, and B3LYP37 lev-
els of theory. The latter two methods have produced onlyD2h structures. For all other
methods, two minima were found, and theC2v-D2h energy differences∆E’s were cal-
culated from the corresponding total energies. In addition,∆E’s were calculated by
a variety of methods at the UHF-CCSD optimized geometries. The lowest electron-
ically excited states forC2v andD2h structures where computed by EOM-EA-CCSD
at the ground state geometries optimized at the same method. We also performed
the UHF-CCSD, UHF-CCSD(T), OO-CCD, OO-CCD(T), EOM-EA-CCSD, EOM-SF-
CCSD and EOM-EA-OD PES scans along theD2h-C2v displacement coordinateR(x)
defined as:
R(x) = (1− x) ∗ r(D2h) + x ∗ r(C2v), (2.2)
wherex is a fraction ofC2v structure and vectorr denotes Cartesian coordinates of the
D2h andC2v structures.
The photodetachment spectra were calculated within double harmonic and parallel
normal modes approximations (using normal modes of anion) by programPES 38. In all
spectra calculations, we employed the SF-DFT geometry and frequencies of the singlet
66
state of the diradical39, while the triplet state was described by CCSD. For the anion, we
considered UHF-CCSD and B3LYP geometries and harmonic vibrational frequencies
of the anion.
Charge distributions were obtained from the natural bond orbital (NBO) analysis40.
Dipole moments were computed relative to the molecular center of mass. The Mulliken
conventions41 for symmetry labels were used: for bothC2v andD2h, the plane of the
molecule is YZ and the Z-axis passes through the two carbons that host the unpaired
electrons. Electronic structure calculations were performed usingQ-Chem42 andACES
II 43.
2.4 Equilibrium structures, charge distributions, and
the D2h→ C2v potential energy profiles at different
levels of theory
This section discusses theD2h andC2v optimized structures, respective energy differ-
ences, PES scans, as well as changes in charge distributions upon symmetry lowering.
As pointed out by Nash and Squires16, most theoretical methods predict two minima on
the PES, withD2h andC2v symmetries. As follows from the MOs (see Fig. 2.1), the
latter structures are characterized by a localized charge, i.e., are of a distonic type.
The optimizedD2h structures are presented in Fig. 2.3. Fig. 2.4 and Fig. 2.5 sum-
marizes differences between theD2h andC2v structures calculated at different levels of
theory. For coupled-cluster methods, the changes in geometries between different levels
of theory (Fig. 2.4, 2.5) are not significant and are smaller for theD2h structure than
for C2v. For example, the variation in bond lengths and angles forD2h structures are
67
H
H
H
H
1.3911.3981.3841.3901.3931.393
1.0931.0961.0921.0931.0931.092
121.9121.8122.0121.9121.6121.5
1.4351.4401.4331.4351.4331.433
116.5116.5117.2116.8116.6116.6
CCSDCCSD(T)B3LYPOO-CCDEA-CCSDSF-CCSD
Figure 2.3:D2h structures optimized at different levels of theory.
below 0.02A and 2, respectively. The differences are larger for theC2v structures, pos-
sibly because of larger spin-contamination of the doublet and triplet UHF references.
TheD2h structures calculated by the CC methods with double substitutions, i.e., CCSD,
OO-CCD, EOM-EA-CCSD, and EOM-SF-CCSD, are very close to each other, which
is reassuring in view of the instability of the doublet UHF reference. The changes in the
bond lengths and angles between CCSD(T) and CCSD are within 0.01A and 1, which
is consistent with typical differences for well behaved molecules (Ref. 44). The largest
change due to (T) is observed for the distance between the radical centers. The B3LYP
structure is considerably different, e.g., the corresponding C1-C4 distance is 0.035A
shorter than the CCSD(T) value. The above trends suggest that the accuracy of the CC
structures is not affected by the HF instabilities and vibronic interactions, and question
the reliability of B3LYP results.
68
B 3 L Y P C C S D O O - C C D C C S D ( T ) E A - C C S D S F - C C S D1 1 2
1 1 4
1 1 6
1 1 8
1 2 0
1 2 2
Angle
, deg
M e t h o d
C 2 v : C 4 D 2 h : C 1 C 2 v : C 1
B 3 L Y P C C S D O O - C C D C C S D ( T ) E A - C C S D S F - C C S D1 . 3 8
1 . 3 9
1 . 4 0
1 . 4 1
1 . 4 2
1 . 4 3
1 . 4 4
1 . 4 5
Bond
length
, A
M e t h o d
D 2 h : C 2 - C 3 C 2 v : C 2 - C 3 C 2 v : C 1 - C 2 D 2 h : C 1 - C 2 C 2 v : C 3 - C 4
Figure 2.4: CCC angles, bonds lengths calculated by different coupled-cluster methodsand B3LYP.
The comparisons between theD2h and C2v structures allows one to quantify the
degree of symmetry lowering at each level of theory. For example, the magnitude of
69
B 3 L Y P C C S D O O - C C D S F - C C S D E A - C C S D C C S D ( T )2 . 8 7
2 . 8 8
2 . 8 9
2 . 9 0
2 . 9 1
Distan
ce, A
M e t h o d
D 2 h : C 1 - C 4 C 2 v : C 1 - C 4
Figure 2.5: The distance between the radical-anion centers calculated by differentcoupled-cluster methods and B3LYP.
C2v distortions can be characterized by the difference between C1-C2 and C3-C4, which
are identical inD2h. As shown in the lower panel of Fig. 2.4, the difference in C1-C2
between theD2h andC2v structures is largest for CCSD and OO-CCSD and is reduced
by the inclusion of triples. At the EOM-EA-CCSD level, it shrinks even further. The C1-
C2 — C2-C3 difference, as well as difference in the C1 and C4 angles, follows the same
trend. Thus, the magnitude of symmetry breaking is smaller for higher-level methods,
or methods that use better references. This, along with the absence of aC2v minimum
at the EOM-SF-CCSD level, suggests that its occurrence is artefactual.
The degree of symmetry breaking can also be quantified by the charge localization,
which gives rise to a non-zero dipole moment shown in Fig. 2.6. The correspond-
ing atomic charges calculated using the NBO procedure are consistent with the dipole
moment changes, except for ROHF, which may be an artifact of the NBO analysis. In
agreement with the trends in structural changes, the charge localization is smaller for
70
- 0 . 4 8
- 0 . 4 4
- 0 . 4 0
- 0 . 3 6
- 0 . 3 2
R O H F - E A - C C S D
q(C1), a
.u
U H F
R O H F
U H F - C C S D
R O H F - C C S D U H F - E A - C C S D
2 . 0
2 . 5
3 . 0
3 . 5
4 . 0
4 . 5
5 . 0
5 . 5
Dipole
, deb
ye
U H FR O H F
U H F - C C S D
R O H F - C C S D
U H F - E A - C C S D
R O H F - E A - C C S D
Figure 2.6: The NBO charge at C1 (top panel) and dipole moments (bottom panel)at the optimizedC2v geometries. The ROHF-CCSD and ROHF-EA-CCSD values arecalculated at the geometries optimized by the corresponding UHF-based method.
71
more correlated wave functions. Interestingly, the overall dipole moment decreases in
spite of increased C1-C4 distance. Thus, the charge localization patterns also suggest
that the symmetry lowering is an artifact of an incomplete treatment of electron correla-
tion.
Finally, symmetry lowering can be characterized by energy differences (∆E’s)
between theD2h andC2v structures summarized in Table 2.1 and Fig. 2.7. In addition,
several PES scans are shown in Fig. 2.8. The first part of the table and the upper panel of
Fig. 2.7 summarize single point calculations at the CCSD optimized structures (MCSCF
and CASPT2N employ the MCSCF geometries and are from Ref. 16), while the second
part and the lower panel of Fig. 2.7 present energy differences calculated using the
respective optimized structures. All∆E’s from Table 2.1 and Fig. 2.7 are calculated
using symmetry-pure UHF references atD2h (see section 2.3). Fig. 2.8 shows both the
symmetry-pure and symmetry-broken CCSD/CCSD(T) values atD2h. For these meth-
ods, the correct-symmetry reference yields lower correlated energies, which results in
cusps on the PESs (see section 2.3). OO-CCD has only one solution atD2h minimum,
and the respective curves are smooth.
Similarly to the trends in structures and charge localization, the energy difference
between theD2h andC2v structures is also very sensitive to correlation treatment, in
agreement with earlier studies16. UHF and ROHF place theC2v structure significantly
lower (by about 1 eV), than theD2h one, while MP2 gives the almost exactly oppo-
site result. At higher level methods, the energy difference becomes much smaller. At
the CCSD and OO-CCD levels, theC2v structure is just a little (<0.1 eV) more stable
than theD2h one. Finally, (T) correlation at the CCSD or OO-CCD levels changes the
balance and theD2h structure becomes the global minimum. Similarly, at the EOM-EA-
CCSD level,D2h is also the lowest minimum. The B3LYP and EOM-SF PES do not
have aC2v minimum. Overall, the PES along this distortion is rather flat (due to vibronic
72
- 1 . 2
- 0 . 8
- 0 . 4
0 . 0
0 . 4
0 . 8
1 . 2
UHF-E
A-OD
OO-C
CDOO
-CCD
(T)
ROHF
-CCS
D(T)
E(C2v
) - E(D
2h), e
V
UHF
ROHF
UHF-M
P2RO
HF-M
P2UH
F-CCS
DRO
HF-C
CSD
DFT/B
3LYP
UHF-C
CSD(
T)
UHF-E
A-CCS
DUH
F-SF-C
CSD
MCSC
FCA
SPT2
N
- 1 . 2
- 0 . 8
- 0 . 4
0 . 0
0 . 4
0 . 8
1 . 2
OO-C
CD
E(C2v
) - E(D
2h), e
V
UHF
ROHF
UHF-C
CSD
UHF-C
CSD(
T)
UHF-E
A-CCS
D
Figure 2.7: Energy differences between theC2v andD2h structures (6-311+G** basis).Upper panel: calculated using the UHF-CCSD optimized geometries. MCSCF andCASPT2N employ the MCSCF geometries. Lower panel: calculated at the respectiveoptimized geometries.
73
Table 2.1: Energy differences between theC2v andD2h minima (eV) in the 6-311+G**basis set. Negative values correspond to theC2v minimum being lower. ZPEs are notincluded.
Method UHF ROHF
at UHF-CCSD optimized geometries
HF -0.999 -1.049
MP2 1.191 0.887
CCSDa -0.046 (-0.116) -0.130
CCSD(T)a 0.140 (-0.052) 0.077
B3LYP 0.070
5050b -0.149
OO-CCD -0.096
OO-CCD(T) 0.081
EOM-EA-CCSD 0.086
EOM-EA-OD 0.050
EOM-SF-CCSD 0.052
at the geometries optimized by the corresponding method
HF -0.971 -1.039
5050 -0.136
CCSD -0.046
CCSD(T) 0.140
OO-CCD -0.097
EOM-EA-CCSD 0.071
a Calculated using symmetry-pure and symmetry-broken (values in parenthesis) doubletUHF references atD2h.b DFT functional with 50% of the HF exchange.
interactions): the most reliable estimates of∆Es between the two CCSD optimized min-
ima range between 0.05 eV (EOM-SF-CCSD) to 0.08 eV [CCSD(T) and OO-CCD(T)].
Optimized-orbital and regular CC methods give similar energy orderings, although OO-
CCD and OO-CCD(T) place theD2h structure a little higher than the corresponding CC
methods based on the Hartree-Fock reference. EOM-SF behave similarly to EOM-EA.
74
0 2 0 4 0 6 0 8 0 1 0 0- 0 . 1 6
- 0 . 1 2
- 0 . 0 8
- 0 . 0 4
0 . 0 0
0 . 0 4
0 . 0 8
0 . 1 2
E-E
(C2v
), eV
% o f C 2 v
C C S D C C S D ( T ) O O - C C D O O - C C D ( T )
E A - C C S D S F - C C S D E A - O D B 3 L Y P
Figure 2.8: The potential energy profiles along theD2h-C2v scans. At theD2h struc-ture, CCSD and CCSD(T) energies are calculated using symmetry-broken and pure-symmetry UHF references (see section 2.3).
Similar trends, although with larger variations, were observed for multi-reference
methods, e.g. MCSCF and CASPT2. The previous study16 reported the energy dif-
ferences between theC2v andD2h structures calculated by multi-reference methods as
-0.286 eV for MCSCF(9,8)/cc-pVDZ and 0.388 for CASPT2N/cc-pVDZ. Thus, only
when dynamical correlation is included in a multi-reference method, theD2h structure
becomes the lowest in energy.
As follows from Table 2.1, using different optimized geometries has only a minor
effect on∆E’s. For example, the energy differences calculated at the CCSD optimized
geometries are within 0.03 eV from those calculated using the structures optimized at
the same level of theory, and the observed trends are the same.
75
Thus, the presented∆E’s exhibit the same trend as the optimized geometries and
charge localization patterns — the more correlation we add the more stable theD2h
structure becomes.
Energy profiles from Fig. 2.8 give more detailed information on the shape of the
PES. Due to the HF instabilities, the CCSD and CCSD(T) curves are discontinuous,
unless the symmetry-broken UHF reference is used atD2h. Apart from the cusp, the
shapes of the OO-CCD and CCSD curves are surprisingly similar. The difference
becomes more profound when triples are included: the OO-CCD(T) curve smoothly
descends towards theD2h minimum, while the CCSD(T) curve is gradually climbing up
all the way till the final drop atD2h. Even though the correspondingD2h→ C2v ∆E’s
are close, the shape of the EOM-SF-CCSD and OO-CCD(T) curves are quite different:
the former is much flatter atD2h than the latter. The EOM-EA-CCSD curve parallels
EOM-SF-CCSD aroundD2h, but soon becomes spoiled by the instability in the triplet
reference. Using optimized orbitals triplet reference solves the problem of HF instabil-
ity and the EOM-EA-OD curve is very similar to the EOM-SF-CCSD one. The B3LYP
curve is steeper than the EOM-SF one, consistent with large value of the corresponding
frequency (see next section). Overall, Fig. 2.8 demonstrates that describing the shape of
PES is more challenging than calculating energy difference between two minima. One
should expect considerable differences in the corresponding vibrational frequency. Note
that the analytic CCSD or CCSD(T) frequencies would differ dramatically from those
calculated by finite differences: the analytic frequencies will characterize the curvature
of the PES corresponding to the symmetry-pure reference, whereas finite difference cal-
culations will sample the cusp.
To conclude, the analysis of the equilibrium structures, charge distributions, and
energy differences calculated at different levels of theory suggests that theC2v minimum
is an artifact of incomplete treatment of electron correlation and HF instabilities in the
76
doublet or triplet references. The strong vibronic interactions result in rather flat PES
along theD2h→ C2v distortion. In the next section, we present our calculations of the
photoelectron spectrum, which allows us to assess the quality of calculated geometries
and to unambiguously determine the symmetry of the anion.
2.5 Vibrational frequencies and the photoelectron spec-
trum of para-C 6H−4
Photoelectron spectroscopy is a very sensitive structural tool. The spectra depend on
equilibrium geometries, frequencies and normal modes of the initial (radical anion) and
target (diradical) states. The equilibrium geometries were discussed in the previous sec-
tion. The vibrational frequencies of the anion (atD2h) and the diradical are given in
Table 2.2. The CCSD and B3LYP frequencies of the anion appear to be similar, except
for the lowestb1u mode that describesD2h→C2v displacements and is most affected by
vibronic interactions. This mode also exhibits the largest change relative to the neu-
tral, i.e., it is considerably softer in C6H−4 . The B3LYP frequency is 40% higher than
the CCSD value, in favor of our assumption that B3LYP underestimates the magni-
tude of vibronic interaction. Unfortunately, the only significant vibrational progressions
present the Franck-Condon factors (see section 2.3) are those for fully-symmetric nor-
mal modes that exhibit displacements relative to the anion. Modes with significant fre-
quency change, such as the aboveb1u mode, yield much smaller features that are likely
to be buried under major progressions, as it happens for the para-benzyne anion.
The electron photodetachment spectrum of the para-benzyne radical anion reported
by Wenthold and coworkers12 consists of the two overlapping bands corresponding to
the singlet (X1Ag) and triplet (3Bu) states of the diradical with the origins, i.e., electron
77
Table 2.2: Vibrational frequencies of the para-benzyne radical anion and the diradical.
Symmetry X2A1a X2A1
b a3B1ua X1A1
c
ag 3137 3085 3221 3317
ag 1471 1434 1565 1440
ag 1189 1175 1166 1202
ag 992 981 1028 1065
ag 624 631 615 660
au 879 917 920 994
au 380 398 368 450
b1g 691 721 810 722
b2g 1120 890 871 961
b2g 647 601 437 616
b3g 3114 3061 3207 3302
b3g 1579 1549 1652 1738
b3g 1297 1284 1300 1327
b3g 608 612 585 597
b1u 3111 3056 3206 3300
b1u 1445 1425 1471 1511
b1u 1071 1060 1046 1101
b1u 452 639 947 973
b2u 3136 3082 3220 3316
b2u 1331 1332 1375 1448
b2u 1189 1203 1267 1254
b2u 1027 1020 1081 1076
b3u 715 718 753 796
b3u 422 447 400 477
a Calculated at UHF-CCSD/6-311+G**b Calculated at B3LYP/6-311+G**c From Ref. 39, SF-5050/6-31G**
affinities (EAs), at 1.265± 0.008 eV and 1.430± 0.015 eV, respectively. Long vibra-
tional progressions for frequencies at 635± 20 cm−1 and 990± 20 cm−1 (assigned to
the singlet band), at 610± 15 cm−1 and 995± 20 cm−1 (assigned to the triplet band),
78
as well as a hotband at 615± 30 cm−1 were reported. For both the singlet and the
triplet bands, the lower (about 600cm−1) and the higher (about 1000cm−1) energy pro-
gressions were assigned to symmetric ring deformation and ring breathing, respectively.
These bands, four lowestag vibrations, are also present in all the computed spectra.
Including the lowestb1u mode resulted in very small features, completely obscured by
the dense lines of major progressions.
In the original paper12, the normal displacements were derived from the fitting pro-
cedure. Table 2.3 and Fig. 2.9 compare these values with the shifts computed in this
work from the correspondingD2h optimized geometries and normal modes of the anion.
The triplet and the singlet states of the diradical are described by the CCSD and spin-flip
DFT methods, respectively (see section 2.3). The computed CCSD shifts are very close
to those derived from the experiment. Values obtained with the CCSD normal modes
and different equilibrium structures [e.g., CCSD(T), EOM-EA-CCSD, OO-CCD, and
B3LYP] are quite similar to each other. However, the full B3LYP shifts (the last line
in Table 2.3), i.e., those computed using the B3LYP normal modes and B3LYP equi-
librium structures, are much larger for the ring breathing mode, which will result in a
longer progression.
Finally, Fig. 2.10 compares the calculated and experimental photodetachment spec-
tra. The upper panel of Fig. 2.10 shows the singlet and triplet bands of the spectrum
for theD2h structures calculated using the CCSD description of the anion. To compare
with the experimental spectrum, the scaled intensities of the singlet and triplet bands
were added together. Two scaling coefficients were determined from the following con-
ditions: (i) the combined spectrum is normalized to 1; and (ii) the intensity of peak
number 3 in the experimental and the calculated spectra coincide. The experimental
spectrum was normalized to 1 as well. The bottom part of Fig. 2.10 compares the
computedD2h and the experimental spectra. The agreement between the two spectra
79
Table 2.3: Displacements in normal modes upon ionization,A ∗√
amu.
Method ν5a ν4
b ν5a ν4
b
Singlet Triplet
Exp. fitc] -0.44 -0.14 0.76 -0.12
CCSDd -0.49 -0.21 -0.69 -0.09
CCSD(T)d -0.50 -0.27 -0.71 -0.14
EOM-EA-CCSDd -0.49 -0.22 -0.70 -0.10
OO-CCDd -0.49 -0.21 -0.69 -0.09
DFT/B3LYPd -0.47 -0.17 -0.67 -0.05
DFT/B3LYPe -0.40 -0.29 -0.60 -0.23
aRing deformationbRing breathingcRef. 12dusing the CCSD normal modeseusing the DFT/B3LYP normal modes
S - Q 6 3 5 S - Q 9 9 0 T - Q 6 1 0 T - Q 9 9 5- 0 . 8
- 0 . 6
- 0 . 4
- 0 . 2
0 . 0 e x p . f i t C C S D C C S D ( T ) E O M - E A - C C S D O O - C C D D F T D F T - f u l l
Displa
ceme
nts, A
∗amu 1/2
N o r m a l m o d e s
Figure 2.9: Displacements along the normal modes upon ionization (D2h structure).
80
1 . 2 1 . 4 1 . 6 1 . 8 2 . 0 2 . 2 2 . 4
Inten
sity
E l e c t r o n b i n d i n g e n e r g y , e V
1 . 2 1 . 4 1 . 6 1 . 8 2 . 0 2 . 2 2 . 4
Inten
sity
E l e c t r o n b i n d i n g e n e r g y , e V
Figure 2.10: Top: Singlet (solid line) and triplet (dotted line) bands of the electron pho-todetachment spectrum forD2h using the CCSD equilibrium geometries and frequenciesof the anion. Bottom: the experimental (solid line) and the calculated (dotted line) spec-tra forD2h using the CCSD structure and normal modes.
81
is excellent: the computed spectrum has the same features as the experimental one and
differs only slightly in the peaks’ intensities. Fig. 2.11 shows the spectrum calculated
for theC2v CCSD structure, which is markedly different — it features additional pro-
gressions that are not present in the experimental spectrum. They correspond to the
ring deformation band and the combination bands involving this mode. To conclude,
an excellent agreement between the spectrum calculated using theD2h structure and
the experimental one, along with significant differences betweenD2h andC2v, unam-
biguously proves theD2h structure of the para-benzyne radical anion. Moreover, it also
confirms high quality of the CCSD equilibrium structure and frequencies for the anion,
as well as the accuracy of the spin-flip description of the diradical. We also calculated
the spectrum using the B3LYP optimized geometries and frequencies for p-C6H−4 (and
the same structures and frequencies for the diradical, as above). The resulting spectrum
(Fig. 2.12) has significantly different ratio of intensities for the two main bands and
much longer progressions. Thus, even though DFT correctly predicts symmetry of the
para-benzyne radical anion, the overall structure is of poor quality. The reasons for this
are discussed below.
2.6 DFT self-interaction error and vibronic interactions
The analysis of equilibrium structures, charge localization, and energy differences
betweenD2h andC2v structures presented in section 2.4 demonstrated that the magni-
tude of symmetry breaking decreases when higher level methods are employed. More-
over, the PES scans and the analysis of UHF references strongly suggest that the exis-
tence of aC2v minimum is an artifact. Indeed, EOM-SF-CCSD, the only CC method
82
1 . 2 1 . 4 1 . 6 1 . 8 2 . 0 2 . 2 2 . 4
Inten
sity
E l e c t r o n b i n d i n g e n e r g y , e V
Figure 2.11: The experimental (solid line) and the calculated (dotted line) spectra forC2v using the CCSD structure and normal modes.
that employs a well-behaved UHF reference (that is, symmetry-pure, not strongly spin-
contaminated and stable), yields only aD2h minimum and a smooth PES. The EOM-
EA-OD potential energy profile parallels the EOM-SF one. Finally, modeling of the
photoelectron spectrum ruled out theC2v structure. Thus, we conclude that symmetry
breaking in p-C6H−4 is purely artefactual.
DFT/B3LYP appears to be more robust with respect to this artefactual symmetry
breaking as it yields only aD2h structure. The important question iswhy, and whether
or not one should consider DFT a reliable tool for modeling systems with strong vibronic
interactions. Detailed benchmark studies (e.g., Ref. 18) provide many examples of DFT
giving “the right answer” for difficult symmetry breaking systems. Stanton explained
these observations by making connections between DFT response theory, TD-DFT exci-
tation energies and their Tamm-Dancoff approximation (footnote 23 in Ref. 20), and
83
1 . 2 1 . 4 1 . 6 1 . 8 2 . 0 2 . 2 2 . 4
Inten
sity
E l e c t r o n b i n d i n g e n e r g y , e V
Figure 2.12: The experimental (solid line) and the calculated (dotted line) spectra forD2h using the B3LYP structure and normal modes.
concluded that DFT describes pseudo Jahn-Teller effects reasonably well “for the right
reason”, that is, because the poles appear in approximately the right places. Other
examples, however, suggested that DFT might actually underestimate the magnitude
of vibronic interactions21. To complicate matters even further, we would like to mention
the dehydro-meta-xylylene anion (DMX−) for which DFT yields low symmetry (Cs or
C2) equilibrium geometries26, while wave function based methods predict planar C2v
structures45.
To gain insight, we begin by analyzing the character of interacting electronic states
in p-C6H−4 and DMX− at symmetric and symmetry-broken geometries. As described in
sections 2.2 and 2.4, the two interacting states of p-C6H−4 have the excess charge equally
delocalized between two radical centers (C1 and C4) at D2h, whereasC2v distortions
84
result in the charge localization at one of the carbons. DMX− shows exactly the oppo-
site behavior26,45: at the planarC2v geometries, most of the excess charge resides on the
σ-radical center, whereas C2 or Cs displacements facilitate the flow of the charge into
theπ-system. Thus, in both cases B3LYP favors the structures with more delocalized
excess charge, which immediately brings about the infamous H+2 example46. While HF
dissociation curve for a one-electron H+2 is exact, many DFT methods result in quasi-
bound potential energy curves and with errors as large as 50 kcal/mol! This happens
because DFT over-stabilizes the solutions with an electron being equally split between
the two hydrogens, which is an artifact of self-interaction error (SIE) present in many
functionals47–49. SIE is considerably larger in systems with odd number of electrons.
Self-interaction energy, Coulomb interaction of an electron with itself, is always posi-
tive, and electron delocalization lowers its magnitude thus causing artificial stabilization
of delocalized configurations. Other well understood examples of artificial stabilization
of delocalized states due to SIE include underestimated barriers for radical reactions,
dissociation of cationic radicals, and mixed-valence transition metal dimers49.
As shown in Ref. 48, the magnitude of SIE in H+2 strongly depends on the distance
between the atoms, varying from a negligibly small value around the equilibrium to 55
kcal/mol (B3LYP) at the dissociation limit. The charge-bearing radical centers in the
para-benzyne radical anion are 2.9A apart. At this distance, the H+2 curve exhibits as
much as 60% of its maximum SIE. This suggests substantial SIE in the para-benzyne
radical anion.
To test this assumption, we computed energy differences between theD2h andC2v
minima (using the CCSD optimized geometries) employing functionals with a varying
fraction of HF exchange. The results are shown in Fig. 2.13. The trend is clear: the
relative stabilization of theD2h structure is inversely proportional to the fraction of the
HF exchange in the functional, and, consequently, is proportional to SIE. Combining
85
different correlation functionals with the fixed fraction of the HF exchange has little
effect on the∆E’s, for example, all functionals with 0% HF exchange yield approxi-
matly the sameD2h-C2v enegry separition. Therefore, we conclude that for DFT, the
strengths of vibronic interactions and, consequently, a relative ordering of theD2h and
C2v structures, are governed by SIE. Thus, B3LYP yields the right structure for the
“wrong reason”. Therefore, it is not at all surprising that the quality of the B3LYP
structure and the shape of PES is relatively poor, as follows from the comparison of
the computed photo-electron spectrum with the experimental one. In DMX−, B3LYP
yields the wrong structure, but for the same “right reason” (SIE). Therefore, B3LYP (or
any functional with SIE) is not a reliable method for the systems with strong vibronic
interactions that affect charge localization patters, or when the interacting states are
characterized by considerably different charge distributions.
0 2 0 4 0 6 0 8 0 1 0 0- 1 . 0
- 0 . 8
- 0 . 6
- 0 . 4
- 0 . 2
0 . 0
0 . 2
H FV W NL Y P
7 5 - 2 5
5 0 - 5 0B 3 L Y PB 3 L Y P 5B 3 P W 9 1
E D F 2
E D F 1G i l lB L Y PS V W NB V W NG G 9 9
E(C2v
) - E(D
2h), e
V
H F e x c h a n g e , %
Figure 2.13: Energy differences between theD2h andC2v minima (using the CCSDstructures) calculated using different density functionals.
86
2.7 Calculation of electron affinities of p-benzyne
Different strategies for calculating electron affinities in problematic open-shell systems
were discussed in Ref. 26. The simplest ”brute-force” approach, i.e., to take the differ-
ence between the total energies of the anion and the neutral calculated at the same level
of theory, can give accurate values only if the errors in the corresponding total energies
are very small (which can be achieved only at very high levels of theory), or if both
species are described on the equal footing by the method and errors in the total ener-
gies cancel out. The latter is difficult to achieve in the case when both the neutral and
anion wave functions have a complicated open-shell character. For example, the singlet
state of the diradical is significantly multi-configurational and it would be described less
accurately than the anion by single reference methods. The triplet state of the diradical,
however, is single-configurational and can be described by single-reference methods
with approximately the same accuracy as the doublet. Thus, triplet EAs calculated by
energy differences employing correlated single-reference methods are more reliable than
singlet EAs. The latter can be accurately computed by subtracting the experimental (or
a calculated by an appropriate method) singlet—triplet gap from the calculated triplet
electron affinity. This approach was used for calculating EAs of triradicals26,45 and
diradicals50. Below we demonstrate that this scheme in the spirit of isodesmic reactions
indeed yields accurate singlet EAs.
EAs calculated at different levels of theory for the singlet and the triplet states of the
diradical are presented in Table 2.4 and Fig. 2.14. The first two columns present triplet
and singlet EAs computed by energy differences. The last column presents singlet EA
calculated from the corresponding triplet EA and the best theoretical estimate of the
singlet-triplet gap. For the triplet state, which is predominantly single-configurational,
even relatively low level methods using a moderate basis, e.g. B3LYP and MP2 in 6-
311+G**, yield reasonable EAs, e.g., within 0.11 eV and 0.07 eV from the experiment,
87
respectively. For the singlet state, however, EAs calculated by DFT and MP2 are 1.03 eV
and 0.67 eV off, because these methods are not appropriate for the multi-configurational
wave function of the singlet. Note that such brute-force approach reverses the states
ordering in the diradical even at the CCSD level. For both states coupled-cluster meth-
ods, especially with triples’ corrections, are in a reasonable agreement with the exper-
imental values. For example, ROHF-CCSD(T) is within 0.16 eV for triplet and within
0.05 eV for singlet.
Table 2.4: Electron affinities (eV) of the a3B1u and X1Ag states of the para-benzynediradical.
Method Triplet EA Singlet EA Singlet EAa
Experiment 1.430 1.265
Experiment-ZPE 1.375 1.140
6-311+G** basis
B3LYP 1.481 2.172 1.310
UHF -0.994 2.679 -1.165
UHF-MP2 1.547 0.467 1.376
UHF-CCSD 0.893 1.659 0.722
UHF-CCSD(T) 1.207 1.085 1.036
ROHF -1.148 2.356 -1.319
ROHF-MP2 2.053 0.914 1.882
ROHF-CCSD 0.872 1.629 0.701
ROHF-CCSD(T) 1.216 1.094 1.045
OO-CCD 0.875 1.644 0.704
OO-CCD(T) 1.210 1.080 1.039
aug-cc-pVTZ basis
ROHF -1.199 2.297 -1.370
ROHF-MP2 2.266 1.010 2.095
ROHF-CCSD 0.998 1.766 0.827
ROHF-CCSD(T) 1.395 1.258 1.224
aCalculated by using triplet EA and the singlet-triplet gap from Ref. 51.
88
0 . 5
1 . 0
1 . 5
2 . 0
2 . 5 6 - 3 1 1 + G * * a u g - c c - p V T Z
ROHF
-CCS
DRO
HF-C
CSD(
T)
EA, e
V
UHF
DFT/B
3LYP
UHF-M
P2
UHF-C
CSD(
T)UH
F-CCS
D
OO-C
CD
OO-C
CD(T)
E x p .
- 1 . 0
- 0 . 5
0 . 0
0 . 5
1 . 0
1 . 5
2 . 0 6 - 3 1 1 + G * * a u g - c c - p V T Z
ROHF
-CCS
DRO
HF-C
CSD(
T)
EA, e
V
UHF
DFT/B
3LYP
UHF-M
P2
UHF-C
CSD(
T)UH
F-CCS
D
OO-C
CD
OO-C
CD(T)
E x p .
Figure 2.14: Adiabatic electron affinities calculated by energy differences for the singlet(top panel) and triplet (bottom panel) state of the para-benzyne diradical.
89
For quantitative thermochemical results, larger basis sets are required. We performed
additional coupled-cluster calculations with the aug-cc-pVTZ basis (see Table 2.4). EA
for the triplet calculated by CCSD(T) is within 0.02 eV from the experimental one, as
expected for this method. However, the singlet CCSD(T) value is 0.12 eV off, because
of the poor description of the multi-configurational singlet state by ground-state single-
reference methods.
By employing the best theoretical estimate for the singlet-triplet gap in para-
benzyne, i.e., 0.171 eV from Ref. 51, singlet EA calculated from the CCSD(T) triplet
value is within 0.08 eV from the experiment, as compared to 0.12 eV for the brute-force
approach. As follows from the excellent agreement of triplet EA, the accuracy of singlet
EA can be improved by refining the value of the singlet-triplet gap.
2.8 Conclusions
The shape of ground-state PES of C6H−4 is strongly affected by the vibronic interactions
with a low-lying excited state. The magnitude of vibronic interactions depends cru-
cially on the the energy gap and couplings between the interacting states. Complicated
open-shell character of the wave functions of the interacting states and HF instabilities
challengeab initio methodology: the shape of PES along the vibronic coupling coordi-
nate differs dramatically for different methods. For example, many wave function based
methods predict vibronic interactions to be strong enough to result in a lower-symmetry
C2v minimum, in addition to theD2h structure. EOM-SF-CCSD and B3LYP predict
only a single minimum, however, the curvature of the surface is quite different.
The degree of this symmetry lowering can be quantified by structural parameters
(e.g., the magnitude of nuclear displacements), charge localization and the resulting
dipole moment, as well as energy difference between the two minima. We found that
90
the magnitude of symmetry breaking (as characterized by the above metrics) decreases
when higher-level methods are employed. For example, the dipole moment of theC2v
structure decreases in the HF→CCSD→EOM-EA-CCSD series. Likewise, coupled-
cluster methods with triples corrections as well as EOM-EA/SF-CCSD predict thatD2h
is a lower energy structure. Moreover, EOM-SF predicted onlyD2h minima. Calculation
of the potential energy profiles along theD2h→ C2v distortion along with the stability
analysis of UHF references allowed us to attribute the lower-symmetry minimum to the
incomplete treatment of electron correlation and UHF instabilities.
The quality ofD2h structure was assessed by comparing the computed photoelectron
spectrum against the experiment. The spectrum calculated using the CCSD description
of the anion and the SF-DFT results for the diradical is in an excellent agreement with
the experimental one, which is particularly encouraging in view of its dense and com-
plicated nature. The spectrum calculated using theC2v geometry and frequencies is
markedly different, which allows us to rule out theC2v structure.
As pointed out in earlier studies16, B3LYP appears to be robust w.r.t. this artefactual
symmetry breaking, and yields onlyD2h structure. The analysis of the structure and the
potential energy profile along theD2h→ C2v distortion reveals that the structure and the
curvature are considerably different from those calculated by reliable wave functions
methods. Moreover, the computed spectrum differs significantly from the experimental
one. The shape of the surface, as well as the corresponding harmonic frequency reveals
that B3LYP underestimates the magnitude of vibronic interactions, as pointed out ear-
lier by Crawford and coworkers21. Further analysis allowed us to attribute the failure
of B3LYP to the self-interaction error. A similar behavior (that yielded seemingly dif-
ferent result) was observed in DMX−, where SIE resulted in B3LYP overestimation
of vibronic interactions and, consequently, a lower symmetry structure. In both cases,
91
large SIE originates in significant changes in charge localization patterns along sym-
metry breaking coordinate. Thus, B3LYP is not a reliable method for systems where
vibronic interactions are coupled to different charge localization patterns, as happens in
distonic radical and diradical anions.
The work presented in this chapter was published in Ref. 52.
Reference list
[1] Wenk, H.H. ; Winkler, M. ; Sander, W.Angew. Chem. Int. Ed. Engl.2003, 42, 502.
[2] Jones, R.R. ; Bergman, R.G.J. Am. Chem. Soc.1972, 94, 660.
[3] Lee, M.D. ; Dunne, T.S. ; Siegel, M.M. ; Chang, C.C. ; Morton, G.0. ; Borders,D.B. J. Am. Chem. Soc.1987, 109, 3464.
[4] Borders, D.B. , Doyle, T.W. , Eds.Enediyne antibiotics as antitumor agents;Mar-cel Deekker, New York, 1995.
[5] Maeda, H. , Edo, K. , Ishida, N. , Eds.Neocarzinostatin: the past, present, andfuture of an anticancer drug;Springer, New York, 1997.
[6] Schottelius, M.J. ; Chen, P.Angew. Chem. Int. Ed. Engl.1996, 3, 1478.
[7] Hoffner, J. ; Schottelius, M.J. ; Feichtinger, D. ; Chen, P.J. Am. Chem. Soc.1998,120, 376.
[8] Nicolaou, K.C. ; Smith, A.L.Acc. Chem. Res.1992, 25, 497.
[9] Bowles, D.M. ; Palmer, G.J. ; Landis, C.A. ; Scott, J.L. ; Anthony, J.E.Tetrahedron2001, 57, 3753.
[10] Zeidan, T. ; Manoharan, M. ; Alabugin, I.V.J. Org. Chem.2006, 71, 954.
[11] Wenthold, P.G. ; Hu, J. ; Squires, R.R.J. Am. Chem. Soc.1996, 118, 11865.
[12] Wenthold, P.G. ; Squires, R.R. ; Lineberger, W.C.J. Am. Chem. Soc.1998, 120,5279.
[13] Eshdat, L. ; Berger, H. ; Hopf, H. ; Rabinovitz, M.J. Am. Chem. Soc.2002, 124,3822.
92
[14] Alabugin, I.V. ; Kovalenko, S.V.J. Am. Chem. Soc.2002, 124, 9052.
[15] Alabugin, I.V. ; Manoharan, M.J. Am. Chem. Soc.2003, 125, 4495.
[16] Nash, J.J. ; Squires, R.R.J. Am. Chem. Soc.1996, 118, 11872.
[17] Davidson, E.R. ; Borden, W.T.J. Phys. Chem.1983, 87, 4783.
[18] Cohen, R.D. ; Sherrill, C.D.J. Chem. Phys.2001, 114, 8257.
[19] Crawford, T.D. ; Stanton, J.F. ; Allen, W.D. ; Schaefer III, H.F.J. Chem. Phys.1997, 107, 10626.
[20] Stanton, J.F.J. Chem. Phys.2001, 115, 10382.
[21] Russ, N.J. ; Crawford, T.D. ; Tschumper, G.S.J. Chem. Phys.2004, 120, 7298.
[22] Lowdin, P.-O.Rev. Mod. Phys.1963, 35, 496.
[23] Eisfeld, W. ; Morokuma, K.J. Chem. Phys.2000, 113, 5587.
[24] Crawford, T.D. ; Kraka, E. ; Stanton, J.F. ; Cremer, D.J. Chem. Phys.2001, 114,10638.
[25] Wladyslawski, M. ; Nooijen, M. InACS Symposium Series, Vol. 828, pages 65–92,2002.
[26] Slipchenko, L.V. ; Krylov, A.I.J. Phys. Chem. A2006, 110, 291.
[27] Krishnan, R. ; Binkley, J.S. ; Seeger, R. ; Pople, J.A.J. Chem. Phys.1980, 72, 650.
[28] Frisch, M.J. ; Pople, J.A. ; Binkley, J.S.J. Chem. Phys.1984, 80, 3265.
[29] Woon, D. E. ; Jr., T. H. DunningJ. Chem. Phys.1994, 100.
[30] Cristian, A.M.C. ; Shao, Y. ; Krylov, A.I.J. Phys. Chem. A2004, 108, 6581.
[31] Purvis, G.D. ; Bartlett, R.J.J. Chem. Phys1982, 76, 1910.
[32] Raghavachari, K. ; Trucks, G.W. ; Pople, J.A. ; Head-Gordon, M.Chem. Phys.Lett.1989, 157, 479.
[33] Sherrill, C.D. ; Krylov, A.I. ; Byrd, E.F.C. ; Head-Gordon, M.J. Chem. Phys.1998,109, 4171.
[34] Nooijen, M. ; Bartlett, R.J.J. Chem. Phys.1995, 102, 3629.
[35] Krylov, A.I. Chem. Phys. Lett.2001, 338, 375.
93
[36] Levchenko, S.V. ; Krylov, A.I.J. Chem. Phys.2004, 120, 175.
[37] Becke, A.D.J. Chem. Phys.1993, 98, 5648.
[38] Arnold, D.W. Ph.D. Thesis, UC Berkeley,1994.
[39] Shao, Y. ; Head-Gordon, M. ; Krylov, A.I.J. Chem. Phys.2003, 118, 4807.
[40] NBO 5.0. Glendening, E.D. ; Badenhoop, J.K. ; Reed, A.E. ; Carpenter, J.E. ;Bohmann, J.A. ; Morales, C.M. ; Weinhold, F. Theoretical Chemistry Institute,University of Wisconsin, Madison, WI,2001.
[41] Mulliken, R.S.J. Chem. Phys.1955, 23, 1997.
[42] Shao, Y. ; Molnar, L.F. ; Jung, Y. ; Kussmann, J. ; Ochsenfeld, C. ; Brown, S. ;Gilbert, A.T.B. ; Slipchenko, L.V. ; Levchenko, S.V. ; O’Neil, D. P. ; Distasio Jr.,R.A. ; Lochan, R.C. ; Wang, T. ; Beran, G.J.O. ; Besley, N.A. ; Herbert, J.M. ; Lin,C.Y. ; Van Voorhis, T. ; Chien, S.H. ; Sodt, A. ; Steele, R.P. ; Rassolov, V. A. ;Maslen, P. ; Korambath, P.P. ; Adamson, R.D. ; Austin, B. ; Baker, J. ; Bird, E.F.C.; Daschel, H. ; Doerksen, R.J. ; Drew, A. ; Dunietz, B.D. ; Dutoi, A.D. ; Furlani,T.R. ; Gwaltney, S.R. ; Heyden, A. ; Hirata, S. ; Hsu, C.-P. ; Kedziora, G.S. ;Khalliulin, R.Z. ; Klunziger, P. ; Lee, A.M. ; Liang, W.Z. ; Lotan, I. ; Nair, N. ;Peters, B. ; Proynov, E.I. ; Pieniazek, P.A. ; Rhee, Y.M. ; Ritchie, J. ; Rosta, E. ;Sherrill, C.D. ; Simmonett, A.C. ; Subotnik, J.E. ; Woodcock III, H.L. ; Zhang, W.; Bell, A.T. ; Chakraborty, A.K. ; Chipman, D.M. ; Keil, F.J. ; Warshel, A. ; Herhe,W.J. ; Schaefer III, H.F. ; Kong, J. ; Krylov, A.I. ; Gill, P.M.W. ; Head-Gordon, M.Phys. Chem. Chem. Phys.2006, 8, 3172.
[43] ACES II. Stanton, J.F. ; Gauss, J. ; Watts, J.D. ; Lauderdale, W.J. ; Bartlett, R.J.1993. The package also contains modified versions of the MOLECULE Gaussianintegral program of J. Almlof and P.R. Taylor, the ABACUS integral derivativeprogram written by T.U. Helgaker, H.J.Aa. Jensen, P. Jørgensen and P.R. Taylor,and the PROPS property evaluation integral code of P.R. Taylor.
[44] Helgaker, T. ; Jørgensen, P. ; Olsen, J.Molecular electronic structure theory;Wiley& Sons, 2000.
[45] Munsch, T.E. ; Slipchenko, L.V. ; Krylov, A.I. ; Wenthold, P.G.J. Org. Chem.2004, 69, 5735.
[46] Bally, T. ; Sastry, G.N.J. Phys. Chem. A1997, 101, 7923.
[47] Polo, V. ; Kraka, E. ; Cremer, D.Molecular Physics2002, 100, 1771 .
[48] Zhang, Y. ; Yang, W.J. Chem. Phys.1998, 109, 2604.
94
[49] Lundber, M. ; Siegbahn, P.E.M.J. Chem. Phys.2005, 122, 1.
[50] Reed, D.R. ; Hare, M. ; Kass, S.R.J. Am. Chem. Soc.2000, 122, 10689.
[51] Slipchenko, L.V. ; Krylov, A.I.J. Chem. Phys.2002, 117, 4694.
[52] Vanovschi, V. ; Krylov, A.I. ; Wenthold, P.G.Theor. Chim. Acta2008, 120, 45.
95
Chapter 3
Future work
The effective fragment potential method, which emerged in the last decade, is currently
an active area of research in both the methodology and applications. The implementation
developed in this work increases the availability of the method and opens the possibil-
ity to combine it with variousab initio theories available in Q-Chem1 (e.g., RI-CCSD,
EOM-SF-CCSD). Thus, it is likely to increase the interest to this method and its poten-
tial even further. Since EFP is a relatively new method, the opportunities for the future
developments are abound. Below the most interesting directions of the future work are
presented in the separate sections for the method, implementation and applications.
3.1 Methodology
Three-body dispersion
A recent work by Podeszwa ar al.2 demonstrated that three-body dispersion can con-
tribute up to 50% to total three-body interaction energy for non-polar molecular clus-
ters, such as cyclic benzene timer. Therefore, we expect that three body dispersion will
become increasingly important in the three-dimensional aromatic clusters and in the
bulk. To model intermolecular interactions in such systems accurately, the three-body
dispersion term needs to be included in the EFP model. The approach, which must not
compromise the computational efficiency of EFP, is yet to be developed. One of the
96
possible directions is to apply the approximations made in the derivation of the two-
body EFP dispersion3 to the three-body dispersion expression in terms of frequency-
dependent electric polarizabilities derived in Ref. 4.
Impoved electrostatics expansion
In the present EFP formulation5, the exchange-repulsion, polarization and dispersion
points are the LMO centroids, whereas the electrostatics expansion points are the nuclei
and bond midpoints. The LMOs centroids account for both bond midpoints and lone
pairs. Thus, the lone pairs, which also represent concentration of the electronic density,
are not included in the electrostatic expansion. An interesting question is if using the
LMO centroids and the nuclei (instead of the bond midpoints and the nuclei) as the
electrostatics expansion points will make the EFP description more balanced and thus
more accurate. Additionally, the multipoles at the LMO centroids might better fit in the
unified EFP damping scheme proposed in Ref. 29.
Linear scaling EFP
The computational complexity of all EFP components in their straightforward formula-
tion is O(N2), whereN is the number of fragments. Application of the fast multipole
method6 to the electrostatics and polarization components along with a cutoff procedure
to fast decaying exchange-repulsion and dispersion, will bring the computational com-
plexity down toO(Nlog(N)). This will open the possibility for modeling much larger
molecular systems and the properties of the bulk condensed phase at the EFP level.
Correlated treatment of the active region
In the present EFP formulation5 the active region is treated at the HF level of theory,
which does not account for electron correlation. However, it is well established7–18 that
97
accurate description of many molecular properties requires correlated methods. Elec-
tron correlation is especially important for the open-shell species, such as multiconfigu-
rational excited states, which require inclusion of at least static correlation.
The implementation of EFP developed in this work is a part of Q-Chem. This state
of the artab initio electronic structure package implements a wide variety of correlated
methods1. Therefore, a promising direction of the future work is to develop a correlated
EFP theory and to extend our implementation so that the active region can be treated
with any of correlation methods available in Q-Chem. This theory will allow one to
account for the environmental effects in the electronic structure studies of radicals and
electronically excited species, as well as to model chemical reactions in the condensed
phase. Relevant to the research conducted in this work (chapter 2), correlated EFP will
allow to model the reaction of Bergman cyclization in solutions.
Below a possible framework for such correlated EFP theory is outlined. In the
present EFP formulation, the effective fragments affect the active region via one-electron
terms in the Hamiltonian. These terms originate from the electrostatics and polariza-
tion components. Correlated methods can similarly use this modified Hamiltonian to
account for the field of the effective fragments. However, a complication arises from
the polarization component. Its contribution to the Hamiltonian depends on the values
of the induced dipoles. The induced dipoles depend on the wavefunction of the active
region, which in turn depends on the Hamiltonian. Thus the correlated wavefunction
and the induced dipoles should be consistent with each other. To satisfy the consistency
requirement, the induced dipoles and the correlated wavefunctions should be computed
self-consistently with the following iterative procedure:
1. Compute the correlated wavefunction for a given values of the induced dipoles.
2. Update the induced dipoles to account for the new wavefunction.
98
3. If the induced dipoles and the wavefunction are converged exit, otherwise return
to step 1.
However, such strict self-consistency requires to recompute the correlated wavefunc-
tion multiple times and, therefore, is computationally expensive. Alternatively, one may
use the induced dipoles consistent with the HF wavefunction, which is computed as a
prerequisite for all single reference correlated methods. This way the consistency is
compromised to reduce the computational cost. Yet another approach is to perform just
one step toward the consistency, i.e., to compute the correlated wavefunction based on
the HF induced dipoles, update the induced dipoles and recompute the correlated wave-
function again. The accuracy of the last two approximations is yet to be investigated.
For the methods computing the correlated wavefunction iteratively (such as CCSD), it
is possible to design a more efficient integrated approach, similar to that for the Hartree-
Fock SCF procedure (section 1.3). In this approach, at every iteration both the correlated
wavefunction and the induced dipoles are updated and thus consistence of both will be
achieved once they converge. The drawback is the increased complexity in the imple-
mentation.
3.2 Implementation
Gradients
As discussed in section 1.7, the energy minima for the benzene dimer structures were
obtained using scans of the potential energy surface. For large systems, a better approach
would be to use a gradient-based optimization procedure. The gradient of the electro-
static energy was formulated in the original EFP paper19. Gradient for the dispersion
energy is simply a derivative of1R6 . More intricate and difficult to implement are recently
99
published gradients for polarization20,21and exchange-repulsion energies22. Implement-
ing the gradient of EFP energy will allow one to perform geometry optimizations and
molecular dynamics simulations for large molecular systems.
Fragment paramters
In the developed implementation, the EFP fragment parameters are defined in the Q-
Chem standard input, as described in section 1.6. The parameters of an effective frag-
ment are computed from an individual isolated molecules, as discussed in sections 1.2,
1.3 1.4 and 1.5. Thus, for a given molecule the same set of EFP parameters can be used
in all computations. This immediately prompts creating a library of the pre-computed
EFP parameters for molecules that frequently occur in the systems studied with EFP.
Such library can, for instance, include the parameters for common organic solvents
(e.g., water, alcohols, acetone, DMSO, acetonitrile, formic acid, dichloromethane, car-
bon tetrachloride, chloroform, tetrahydrofuran, benzene, phenol, toluene, etc.). The EFP
parameters included in the library can be carefully selected and benchmarked against
high levelab initio theories. Therefore, the benefits of such a library are not just the
convenience it provides, but also a high-quality assurance for the EFP parameters.
QM-EF dispersion and exchange-repulsion interactions
In the current state of EFP theory, the interaction between the effective fragments and the
ab initio region is due to the electrostatics and polarization components only. Thus, there
is no account for dispersion and exchange-repulsion interactions between them. The
proper theories for such interactions are presently developed in M.S. Gordon’s group.
Meanwhile, it is possible to account for such interactions by treating theab initio system
as an effective fragment for these two types of interactions. Since the wavefunction of
theab initio part is readily available, the dispersion and exchange-repulsion parameters
100
can be computed as discussed in sections 1.4 and 1.5, respectively. From these param-
eters, dispersion and exchange-repulsion interaction energies between the active region
and the effective fragments can be evaluated. Implementing such scheme will allow one
to properly compute the binding energy in the systems with anab initio region. In the
present EFP theory, binding energy is properly computed for systems composed of the
effective fragments only, whereas molecular systems with anab initio region are used
to model the solvation effects. Implementing the suggested technique will allow one to
combine accurate description of the solvation effects and binding energies.
3.3 Applications
Numerous applications of EFP for modeling the solvation effects and the binding ener-
gies in the clusters and condensed phase can be envisioned. Here, the future direction of
the research of multipleπ − π interactions conducted in this work (section 1.7) is pre-
sented. As demonstrated in Ref. 2 using the cyclic benzene trimer example, three-body
dispersion has a substantial contribution to many-body interaction energy. Introduction
of the three-body dispersion term in EFP is proposed in section 3.1. Once this term is
available, the extended EFP method can be applied to:
• study the role of many-bodyπ− π interactions in the multi dimensional aromatic
clusters and the bulk;
• model the mechanical properties of carbon-based materials, such as graphite and
fullerene powders;
• investigate the role ofπ − π stacking in the rigidity of DNA.
101
Reference list
[1] Shao, Y. ; Molnar, L.F. ; Jung, Y. ; Kussmann, J. ; Ochsenfeld, C. ; Brown, S. ;Gilbert, A.T.B. ; Slipchenko, L.V. ; Levchenko, S.V. ; O’Neil, D. P. ; Distasio Jr.,R.A. ; Lochan, R.C. ; Wang, T. ; Beran, G.J.O. ; Besley, N.A. ; Herbert, J.M. ; Lin,C.Y. ; Van Voorhis, T. ; Chien, S.H. ; Sodt, A. ; Steele, R.P. ; Rassolov, V. A. ;Maslen, P. ; Korambath, P.P. ; Adamson, R.D. ; Austin, B. ; Baker, J. ; Bird, E.F.C.; Daschel, H. ; Doerksen, R.J. ; Drew, A. ; Dunietz, B.D. ; Dutoi, A.D. ; Furlani,T.R. ; Gwaltney, S.R. ; Heyden, A. ; Hirata, S. ; Hsu, C.-P. ; Kedziora, G.S. ;Khalliulin, R.Z. ; Klunziger, P. ; Lee, A.M. ; Liang, W.Z. ; Lotan, I. ; Nair, N. ;Peters, B. ; Proynov, E.I. ; Pieniazek, P.A. ; Rhee, Y.M. ; Ritchie, J. ; Rosta, E. ;Sherrill, C.D. ; Simmonett, A.C. ; Subotnik, J.E. ; Woodcock III, H.L. ; Zhang, W.; Bell, A.T. ; Chakraborty, A.K. ; Chipman, D.M. ; Keil, F.J. ; Warshel, A. ; Herhe,W.J. ; Schaefer III, H.F. ; Kong, J. ; Krylov, A.I. ; Gill, P.M.W. ; Head-Gordon, M.Phys. Chem. Chem. Phys.2006, 8, 3172.
[2] Podeszwa, R. ; Szalewicz, K.J. Chem. Phys.2007, 126, 194101.
[3] Adamovix, I. ; Gordon, M.S.Mol. Phys.2005, 103, 379.
[4] McLachlan, A.D.Mol. Phys.1963, 6, 423.
[5] Gordon, M.S. ; Slipchenko, L. ; Li, H. ; Jensen, J.H. InAnnual Reports in Com-putational Chemistry; Spellmeyer, D.C. , Wheeler, R. , Eds., Vol. 3 ofAnnualReports in Computational Chemistry;Elsevier, 2007; pages 177 – 193.
[6] Schmidt, K.E. ; Lee, M.A.J. Stat. Phys.1991, 63, 1223.
[7] Woon, D.E. ; Dunning, T.H.J. Chem. Phys.1993, 99, 1914.
[8] Peterson, K.A. ; Kendall, R.A. ; Dunning, T.H.J. Chem. Phys.1993, 99, 1930.
[9] Peterson, K.A. ; Kendall, R.A. ; Dunning, T.H.J. Chem. Phys.1993, 99, 9790.
[10] Peterson, K.A. ; Woon, D.E. ; Dunning, T.H.J. Chem. Phys.1994, 100, 7410.
[11] Woon, D.E.J. Chem. Phys.1994, 100, 2838.
[12] Woon, D.E. ; Dunning, T.H.J. Chem. Phys.1994, 101, 8877.
[13] Peterson, K.A. ; Dunning, T.H.J. Chem. Phys.1995, 102, 2032.
[14] Peterson, K.A. ; Dunning, T.H.J. Chem. Phys.1997, 106, 4119.
[15] Woon, D.E. ; Peterson, K.A. ; Dunning, T.H.J. Chem. Phys.1998, 109, 2233.
102
[16] Wilson, A.K. ; Dunning, T.H.J. Chem. Phys.1997, 106, 8718.
[17] Peterson, K.A. ; Dunning, T.H.J. Phys. Chem. A1997, 101, 6280.
[18] Peterson, K.A. ; Wilson, A.K. ; Woon, D.E. ; Dunning, T.H.Theor. Chem. Acc.1997, 97, 251.
[19] Day, P.N. ; Jensen, J.H. ; Gordon, M.S. ; Webb, S.P. ; Stevens, W.J. ; M.Krauss;D.Garmer; H.Basch; D.CohenJ. Chem. Phys.1996, 105, 1968.
[20] Li, H. ; Gordon, M.S.J. Chem. Phys.2007, 126.
[21] Li, H. ; Netzloff, H.M. ; Gordon, M.S.J. Chem. Phys.2006, 125.
[22] Li, H. ; Gordon, M.S.Theor. Chem. Acc.2006, 115, 385.
103
Bibliography
[1] I. Adamovic and M.S. Gordon. Solvent effects on the S(N)2 reaction: Applicationof the density functional theory-based effective fragment potential method.J.Phys. Chem. A, 109(8):1629–1636, 2005.
[2] I. Adamovix and M.S. Gordon. Dynamic polarizability, dispersion coefficient C6and dispersion energy in the effective fragment potential method.Mol. Phys.,103(2-3):379–387, 2005.
[3] I.V. Alabugin and S.V. Kovalenko. C1-C5 photochemical cyclization ofenediynes.J. Am. Chem. Soc., 124:9052–9053, 2002.
[4] I.V. Alabugin and M. Manoharan. Radical-anionic cyclizations of enediynes:Remarkable effects of benzannelation and remote substituents on cycliorearoma-tization reactions.J. Am. Chem. Soc., 125:4495–4509, 2003.
[5] N.L. Allunger, Y.H. Yuh, and J.H. Lii. Molecular mechanics - the MM3 force-field for hydrocarbons .1.J. Am. Chem. Soc., 111(23):8551–8566, 1989.
[6] D.W. Arnold. Study of Radicals, Clusters, and Transition State Species by AnionPhotoelectron Spectroscopy. PhD thesis, UC Berkeley, 1994.
[7] B. Askew, P. Ballester, C. Buhr, K.S. Jeong, S. Jones, K. Parris, K. Williams, andJ. Rebek. Molecular recognition with convergent functional groups. VI. Syntheticand structural studies with a model receptor for nucleic acid components.J. Am.Chem. Soc., 111(3):1082–1090, 1989.
[8] R. Balawender, B. Safi, and P. Geerlings. Solvent effect on the global and atomicDFT-based reactivity descriptors using the effective fragment potential model.Solvation of ammonia.J. Phys. Chem. A, 105(27):6703–6710, 2001.
[9] T. Bally and G.N. Sastry. Incorrect dissociation behavior of radical ions in densityfunctional calculations.J. Phys. Chem. A, 101:7923–7925, 1997.
[10] A.D. Becke. Density-functional thermochemistry. iii. The role of exact exchange.J. Chem. Phys., 98:5648, 1993.
104
[11] D.B. Borders and T.W. Doyle, editors.Enediyne antibiotics as antitumor agents.Marcel Deekker, New York, 1995.
[12] D.R. Bowler, T. Miyazaki, and M.J. Gillan. Recent progress in linear scaling abinitio electronic structure techniques.J. Phys.: Condens. Matter, 14(11):2781–2798, 2002.
[13] D.M. Bowles, G.J. Palmer, C.A. Landis, J.L. Scott, and J.E. Anthony. TheBergman reaction as a synthetic tool: Advantages and restrictions.Tetrahedron,57:3753–3760, 2001.
[14] S.F. Boys. Construction of some molecular orbitals to be approximately invariantfor changes from one molecule to another.Rev. Mod. Phys., 32(2):296–299, 1960.
[15] S.F. Boys. InQuantum Science of Atoms, Molecules, and Solids, pages 253–262.Academic Press, 1966.
[16] M.F. Brana, M. Cacho, A. Gradillas, B. Pascual-Teresa, and A. Ramos. Interca-lators as anticancer drugs.Curr. Pharm. Des., 7(17):1745–1780, 2001.
[17] A.D. Buckingham. Molecular quadrupole moments.Q. Rev. Chem. Soc.,13(3):183–214, 1959.
[18] S.K. Burley and G.A. Petsko. Aromatic-aromatic interaction: A mechanism ofprotein structure stabilization.Science, 229(4708):23–28, 1985.
[19] C.J. Casewit, K.S. Colwell, and A.K. Rappe. Application of a universal force-field to organic-molecules.J. Am. Chem. Soc., 114(25):10035–10046, 1992.
[20] K.R.S. Chandrakumar, T.K. Ghanty, S.K. Ghosh, and T. Mukherjee. Hydra-tion of uranyl cations: Effective fragment potential approach.J. Molec. Struct.(Theochem), 807(1-3):93–99, 2007.
[21] R.D. Cohen and C.D. Sherrill. The performance of density functional theoryfor equilibrium molecular properties of symmetry breaking molecules.J. Chem.Phys., 114:8257–8269, 2001.
[22] T.D. Crawford, E. Kraka, J.F. Stanton, and D. Cremer. Problematic p-benzyne:Orbital instabilities, biradical character, and broken symmetry.J. Chem. Phys.,114:10638–10650, 2001.
[23] T.D. Crawford, J.F. Stanton, W.D. Allen, and H.F. Schaefer III. Hartree-Fockorbital instability envelopes in highly correlated single-reference wave functions.J. Chem. Phys., 107:10626–10632, 1997.
105
[24] A.M.C. Cristian, Y. Shao, and A.I. Krylov. Bonding patterns in benzene trirad-icals from structural, spectroscopic, and thermochemical perspectives.J. Phys.Chem. A, 108:6581–6588, 2004.
[25] E.R. Davidson and W.T. Borden. Symmetry breaking in polyatomic molecules:Real and artifactual.J. Phys. Chem., 87:4783–4790, 1983.
[26] P.N. Day, J.H. Jensen, M.S. Gordon, S.P. Webb, W.J. Stevens, M.Krauss,D.Garmer, H.Basch, and D.Cohen. An effective fragment method for modelingsolvent effects in quantum mechanical calculations.J. Chem. Phys., 105(5):1968–1986, 1996.
[27] P.N. Day, M.S. Pachter, R.and Gordon, and G.N. Merrill. A study of water clus-ters using the effective fragment potential and Monte Carlo simulated annealing.J. Chem. Phys., 112(5):2063–2073, 2000.
[28] P.N. Day and R. Pachter. A study of aqueous glutamic acid using the effectivefragment potential method.J. Chem. Phys., 107(8):2990–2999, 1997.
[29] G.R. Desiraju and A. Gavezzotti. From molecular to crystal structure; polynu-clear aromatic hydrocarbons.JCS Chem. Comm., pages 621–623, 1989.
[30] W. Eisfeld and K. Morokuma. A detailed study on the symmetry breaking an itseffect on the potential surface of NO3. J. Chem. Phys., 113:5587–5597, 2000.
[31] L. Eshdat, H. Berger, H. Hopf, and M. Rabinovitz. Anionic cyclization of across-conjugated enediyne.J. Am. Chem. Soc., 124:3822–3823, 2002.
[32] P.M. Felker, P.M. Maxton, and M.W. Schaeffer. Nonlinear raman studiesof weakly bound complexes and clusters in molecular beams.Chem. Rev.,94(7):1787–1805, 1994.
[33] T.K. Firman and C.R. Landis. Valence bond concepts applied to the molecularmechanics description of molecular shapes. 4. Transition metals withπ-bonds.J.Am. Chem. Soc., 123(47):11728–11742, 2001.
[34] M.A. Freitag and M.S. Gordon. Using the effective fragment-potential modelto study the solvation of acetic and formic acids.Abstracts of Papers of theAmerican Chemical Society, 218:U511–U511, 1999. 1.
[35] M.A. Freitag, M.S. Gordon, J.H. Jensen, and W.J. Stevens. Evaluation ofcharge penetration between distributed multipolar expansions.J. Chem. Phys.,112(17):7300–7306, 2000.
106
[36] M.J. Frisch, J.A. Pople, and J.S. Binkley. Self-consistent molecular orbitalmethods 25. Supplementary functions for Gaussian basis sets.J. Chem. Phys.,80:3265–3269, 1984.
[37] J.L. Gao and D.G. Truhlar. Quantum mechanical methods for enzyme kinetics.Annu. Rev. Phys. Chem., 53:467–505, 2002.
[38] E. Gazit. Self-assembled peptide nanostructures: the design of molecular buildingblocks and their technological utilization.Chem. Soc. Rev., 36(8):1263–1269,2007.
[39] E.D. Glendening, J.K. Badenhoop, A.E. Reed, J.E. Carpenter, J.A. Bohmann,C.M. Morales, and F. Weinhold. NBO 5.0. Theoretical Chemistry Institute, Uni-versity of Wisconsin, Madison, WI, 2001.
[40] V. Gogonea, L.M. Westerhoff, and Jr. K.M. Merz. Quantum mechanical/quantummechanical methods. I. A divide and conquer strategy for solving the schrodingerequation for large molecular systems using a composite density functional–semiempirical Hamiltonian.J. Chem. Phys., 113(14):5604–5613, 2000.
[41] M.S. Gordon, M.A. Freitag, P. Bandyopadhyay, J.H. Jensen, V. Kairys, and W.J.Stevens. The effective fragment potential method: A QM-based MM approach tomodeling environmental effects in chemistry.J. Phys. Chem. A, 105(2):293–307,2001.
[42] M.S. Gordon, L. Slipchenko, H. Li, and J.H. Jensen. Chapter 10 the effectivefragment potential: A general method for predicting intermolecular interactions.In D.C. Spellmeyer and R. Wheeler, editors,Annual Reports in ComputationalChemistry, volume 3 ofAnnual Reports in Computational Chemistry, pages 177– 193. Elsevier, 2007.
[43] T. Helgaker, P. Jørgensen, and J. Olsen.Molecular electronic structure theory.Wiley & Sons, 2000.
[44] J.R. Hill and J. Sauer. Molecular mechanics potential for silica and zeolitecatalysts based on ab initio calculations. 2. Aluminosilicates.J. Phys. Chem.,99(23):9536–9550, 1995.
[45] J. Hoffner, M.J. Schottelius, D. Feichtinger, and P. Chen.J. Am. Chem. Soc.,120:376, 1998.
[46] C.A. Hunter and J.K.M. Sanders. The nature ofπ− π interactions.J. Am. Chem.Soc., 112(14):5525–5534, 1990.
[47] ISO/IEC. ISO/IEC 14882:2003: Programming languages: C++. Technical report,International Organization for Standardization, Geneva, Switzerland., 2003.
107
[48] J.H. Jensen and M.S. Gordon. An approximate formula for the intermolecularPauli repulsion between closed shell molecules.Mol. Phys., 89(5):1313–1325,1996.
[49] J.H. Jensen and M.S. Gordon. An approximate formula for the intermolecularPauli repulsion between closed shell molecules. II. Application to the effectivefragment potential method.J. Chem. Phys., 108(12):4772–4782, 1998.
[50] R.R. Jones and R.G. Bergman. p-Benzyne. Generation as an intermediate ina thermal isomerization reaction and trapping evidence for the 1,4-benzenediylstructure.J. Am. Chem. Soc., 94:660, 1972.
[51] W.L. Jorgensen and N.A. McDonald. Development of an all-atom force fieldfor heterocycles. Properties of liquid pyridine and diazenes.J. Molec. Struct.(Theochem), 424(1-2):145 – 155, 1998.
[52] K. Kitaura, E. Ikeo, T. Asada, T. Nakano, and M. Uebayasi. Fragment molecu-lar orbital method: An approximate computational method for large molecules.Chem. Phys. Lett., 313(3-4):701 – 706, 1999.
[53] R. Krishnan, J.S. Binkley, R. Seeger, and J.A. Pople. Self-consistent molecularorbital methods. XX. A basis set for correlated wave functions.J. Chem. Phys.,72:650, 1980.
[54] A.I. Krylov. Size-consistent wave functions for bond-breaking: The equation-of-motion spin-flip model.Chem. Phys. Lett., 338:375–384, 2001.
[55] M.D. Lee, T.S. Dunne, M.M. Siegel, C.C. Chang, G.0. Morton, and D.B. Borders.Calichemicins, a novel family of antitumor antibiotics. 1. Chemistry and partialstructure of calichemicin.J. Am. Chem. Soc., 109:3464, 1987.
[56] K.R. Leopold, G.T. Fraser, S.E. Novick, and W. Klemperer. Current themes inmicrowave and infrared spectroscopy of weakly bound complexes.Chem. Rev.,94(7):1807–1827, 1994.
[57] S.V. Levchenko and A.I. Krylov. Equation-of-motion spin-flip coupled-clustermodel with single and double substitutions: Theory and application to cyclobu-tadiene.J. Chem. Phys., 120(1):175–185, 2004.
[58] H. Li and M.S. Gordon. Gradients of the exchange-repulsion energy in the gen-eral effective fragment potential method.Theor. Chem. Acc., 115(5):385–390,2006.
[59] H. Li and M.S. Gordon. Polarization energy gradients in combined quantummechanics, effective fragment potential, and polarizable continuum model calcu-lations.J. Chem. Phys., 126(12), 2007.
108
[60] H. Li, M.S. Gordon, and J.H. Jensen. Charge transfer interaction in the effectivefragment potential method.J. Chem. Phys., 124(21), 2006.
[61] H. Li, H.M. Netzloff, and M.S. Gordon. Gradients of the polarization energy inthe effective fragment potential method.J. Chem. Phys., 125(19), 2006.
[62] H. Lin and D.G. Truhlar. QM/MM: what have we learned, where are we, andwhere do we go from here?Theor. Chem. Acc., 117(2):185–199, 2007.
[63] G. Lippert, J. Hutter, and M. Parrinello. A hybrid gaussian and plane wave densityfunctional scheme.Mol. Phys., 92(3):477–487, 1997.
[64] P.-O. Lowdin. Rev. Mod. Phys., 35:496, 1963.
[65] M. Lundber and P.E.M. Siegbahn. Quantifying the effects of the self-interactionerror in DFT: When do the delocalized states appear?J. Chem. Phys.,122(224103):1–9, 2005.
[66] A.D. MacKerell, D. Bashford, Bellott, R.L. Dunbrack, J.D. Evanseck, M.J. Field,S. Fischer, J. Gao, H. Guo, S. Ha, D. Joseph-McCarthy, L. Kuchnir, K. Kuczera,F.T.K. Lau, C. Mattos, S. Michnick, T. Ngo, D.T. Nguyen, B. Prodhom, W.E.Reiher, B. Roux, M. Schlenkrich, J.C. Smith, R. Stote, J. Straub, M. Watanabe,J. Wiorkiewicz-Kuczera, D. Yin, and M. Karplus. All-atom empirical potentialfor molecular modeling and dynamics studies of proteins.J. Phys. Chem. B,102(18):3586–3616, 1998.
[67] H. Maeda, K. Edo, and N. Ishida, editors.Neocarzinostatin: the past, present,and future of an anticancer drug. Springer, New York, 1997.
[68] S.L. Mayo, B.D. Olafson, and W.A. Goddard. DREIDING - a generic force-fieldfor molecular simulations.J. Phys. Chem., 94(26):8897–8909, 1990.
[69] A.D. McLachlan. Three-body dispersion forces.Mol. Phys., 6(4):423–427, 1963.
[70] G.N. Merrill and M.S. Gordon. Study of small water clusters using the effectivefragment potential model.J. Phys. Chem. A, 102(16):2650–2657, 1998.
[71] G.N. Merrill and S.P. Webb. Anion-water clusters A(-)(H2O)(1-6), A = OH,F, SH, Cl, and Br. an effective fragment potential test case.J. Phys. Chem. A,107(39):7852–7860, 2003.
[72] G.N. Merrill and S.P. Webb. The application of the effective fragment potentialmethod to molecular anion solvation: A study of ten oxyanion-water clusters,A(-)(H2O)(1-4). J. Phys. Chem. A, 108(5):833–839, 2004.
109
[73] G.N. Merrill, S.P. Webb, and D.B. Bivin. Formation of alkali metal/alkaline earthcation water clusters, M(H2O)(1-6), M = Li+, Na+, K+, Mg2+, and Ca2+: Aneffective fragment potential (EFP) case study.J. Phys. Chem. A, 107(3):386–396,2003.
[74] G. Monard and K.M. Merz. Combined quantum mechanical/molecular mechan-ical methodologies applied to biomolecular systems.Acc. Chem. Res.,32(10):904–911, 1999.
[75] K. Mueller-Dethlefs, O. Dopfer, and T.G. Wright. ZEKE spectroscopy of com-plexes and clusters.Chem. Rev., 94(7):1845–1871, 1994.
[76] K. Muller-Dethlefs and P. Hobza. Noncovalent interactions: A challenge forexperiment and theory.Chem. Rev., 100(1):143–168, 2000.
[77] R.S. Mulliken. Electronic population analysis on LCAO-MO molecular wavefunctions. I.J. Chem. Phys., 23(10):1833–1840, 1955.
[78] R.S. Mulliken. Report on notation for the spectra of polyatomic molecules.J.Chem. Phys., 23:1997, 1955.
[79] T.E. Munsch, L.V. Slipchenko, A.I. Krylov, and P.G. Wenthold. Reactivity andstructure of the 5-dehydro-m-xylylene anion.J. Org. Chem., 69:5735–5741,2004.
[80] J.J. Nash and R.R. Squires. Theoretical studies of o-,m-, and p-benzyne negativeions. J. Am. Chem. Soc., 118:11872–11883, 1996.
[81] H.M. Netzloff and M.S. Gordon. The effective fragment potential: Small clustersand radial distribution functions.J. Chem. Phys., 121(6):2711–2714, 2004.
[82] K.C. Nicolaou and A.L. Smith. Molecular design, chemical synthesis, and bio-logical action of enediynes.Acc. Chem. Res., 25:497–503, 1992.
[83] M. Nooijen and R.J. Bartlett. Equation of motion coupled cluster method forelectron attachment.J. Chem. Phys., 102:3629–3647, 1995.
[84] K.A. Peterson and T.H. Dunning. Benchmark calculations with correlated molec-ular wave-functions. 7. Binding-energy and structure of the hf dimer.J. Chem.Phys., 102(5):2032–2041, 1995.
[85] K.A. Peterson and T.H. Dunning. Benchmark calculations with correlated molec-ular wave functions. 11. Energetics of the elementary reactions F+H-2, O+H-2,and H’+HCl. J. Phys. Chem. A, 101(35):6280–6292, 1997.
110
[86] K.A. Peterson and T.H. Dunning. Benchmark calculations with correlated molec-ular wave functions. 8. Bond energies and equilibrium geometries of the chn andc2hn (n=1-4) series.J. Chem. Phys., 106(10):4119–4140, 1997.
[87] K.A. Peterson, R.A. Kendall, and T.H. Dunning. Benchmark calculations withcorrelated molecular wave-functions. 2. Configuration-interaction calculations on1st-row diatomic hydrides.J. Chem. Phys., 99(3):1930–1944, 1993.
[88] K.A. Peterson, R.A. Kendall, and T.H. Dunning. Benchmark calculations withcorrelated molecular wave-functions. 3. Configuration-interaction calculations on1st row homonuclear diatomics.J. Chem. Phys., 99(12):9790–9805, 1993.
[89] K.A. Peterson, A.K. Wilson, D.E. Woon, and T.H. Dunning. Benchmark calcu-lations with correlated molecular wave functions. 12. Core correlation effects onthe homonuclear diatomic molecules b-2-f-2.Theor. Chem. Acc., 97(1-4):251–259, 1997.
[90] K.A. Peterson, D.E. Woon, and T.H. Dunning. Benchmark calculations withcorrelated molecular wave-functions. 4. The classical barrier height of the h+h-2-]h-2+h reaction.J. Chem. Phys., 100(10):7410–7415, 1994.
[91] R. Podeszwa. Comment on ”Beyond the benzene dimer: An investigation of theadditivity of π − π interactions”.J. Phys. Chem. A, 112(37):8884–8885, 2008.
[92] R. Podeszwa, R. Bukowski, and K. Szalewicz. Potential energy surface for thebenzene dimer and perturbational analysis ofπ − π interactions.J. Phys. Chem.A, 110(34):10345–10354, 2006.
[93] R. Podeszwa and K. Szalewicz. Three-body symmetry-adapted perturbationtheory based on kohn-sham description of the monomers.J. Chem. Phys.,126(19):194101, 2007.
[94] V. Polo, E. Kraka, and D. Cremer. Electron correlation and the self-interactionerror of density functional theory.Molecular Physics, 100:1771 – 1790, 2002.
[95] W.H. Press, S.A. Teukolsky, W.T. Vetterling, and B.P. Flannery.NumericalRecipes 3rd Edition: The Art of Scientific Computing. Cambridge UniversityPress, New York, NY, USA, 2007.
[96] G.D. Purvis and R.J. Bartlett. A full coupled-cluster singles and doubles model:The inclusion of disconnected triples.J. Chem. Phys, 76:1910, 1982.
[97] J.R. Rabinowitz, K. Namboodiri, and H. Weinstein. A finite expansion methodfor the calculation and interpretation of molecular electrostatic potentials.Int. J.Quant. Chem., 29(6):1697–1704, 1986.
111
[98] K. Raghavachari, G.W. Trucks, J.A. Pople, and M. Head-Gordon. A fifth-orderperturbation comparison of electron correlation theories.Chem. Phys. Lett.,157:479–483, 1989.
[99] D.R. Reed, M. Hare, and S.R. Kass. Formation of gas-phase dianions and distonicions as a general method for the synthesis of protected reactive intermediates.Energetics of 2,3- and 2,6-dehydronaphthalene.J. Am. Chem. Soc., 122:10689–10696, 2000.
[100] R.C. Rizzo and W.L. Jorgensen. OPLS all-atom model for amines: Resolution ofthe amine hydration problem.J. Am. Chem. Soc., 121(20):4827–4836, 1999.
[101] G.V. Rossum.The Python Language Reference Manual. Network Theory Ltd.,2003.
[102] N.J. Russ, T.D. Crawford, and G.S. Tschumper. Real versus artifactual symmetrybreaking effects in Hartree-Fock, density-functional, and coupled-cluster meth-ods.J. Chem. Phys., 120:7298–7306, 2004.
[103] W. Saenger.Principles of Nucleic Acid Structure. Springer-Verlag: New York,1984.
[104] A. Sanz-Garcia and R.J. Wheatley. Atomic representation of the dispersion inter-action energy.ChemPhysChem, 5(5):801–807, 2003.
[105] K.E. Schmidt and M.A. Lee. Implementing the fast multipole method in threedimensions.J. Stat. Phys., 63(5):1223–1235, 1991.
[106] M.W. Schmidt, K.K. Baldridge, S.T. Elbert J.A. Boatz, M.S. Gordon, S. KosekiJ.H. Jensen, N. Mastunaga, K.A. Nguyen, T.L. Windus S. Su, M. Dupuis, andJ.A. Montgomery. General atomic and molecular electronic structure system.J.Comput. Chem., 14(11):1347–1363, 1993.
[107] M.J. Schottelius and P. Chen.Angew. Chem. Int. Ed. Engl., 3:1478, 1996.
[108] M. Schutz, G. Hetzer, and H. Werner. Low-order scaling local electron correlationmethods. I. Linear scaling local MP2.J. Chem. Phys., 111(13):5691–5705, 1999.
[109] E. Schwegler and M. Challacombe. Linear scaling computation of the Hartree-Fock exchange matrix.J. Chem. Phys., 105(7):2726–2734, 1996.
[110] G.E. Scuseria and P.Y. Ayala. Linear scaling coupled cluster and perturbationtheories in the atomic orbital basis.J. Chem. Phys., 111(18):8330–8343, 1999.
[111] Y. Shao, M. Head-Gordon, and A.I. Krylov. The spin-flip approach within time-dependent density functional theory: Theory and applications to diradicals.J.Chem. Phys., 118:4807–4818, 2003.
112
[112] Y. Shao, L.F. Molnar, Y. Jung, J. Kussmann, C. Ochsenfeld, S. Brown, A.T.B.Gilbert, L.V. Slipchenko, S.V. Levchenko, D. P. O’Neil, R.A. Distasio Jr., R.C.Lochan, T. Wang, G.J.O. Beran, N.A. Besley, J.M. Herbert, C.Y. Lin, T. VanVoorhis, S.H. Chien, A. Sodt, R.P. Steele, V. A. Rassolov, P. Maslen, P.P. Koram-bath, R.D. Adamson, B. Austin, J. Baker, E.F.C. Bird, H. Daschel, R.J. Doerk-sen, A. Drew, B.D. Dunietz, A.D. Dutoi, T.R. Furlani, S.R. Gwaltney, A. Hey-den, S. Hirata, C.-P. Hsu, G.S. Kedziora, R.Z. Khalliulin, P. Klunziger, A.M.Lee, W.Z. Liang, I. Lotan, N. Nair, B. Peters, E.I. Proynov, P.A. Pieniazek, Y.M.Rhee, J. Ritchie, E. Rosta, C.D. Sherrill, A.C. Simmonett, J.E. Subotnik, H.L.Woodcock III, W. Zhang, A.T. Bell, A.K. Chakraborty, D.M. Chipman, F.J. Keil,A. Warshel, W.J. Herhe, H.F. Schaefer III, J. Kong, A.I. Krylov, P.M.W. Gill, andM. Head-Gordon. Advances in methods and algorithms in a modern quantumchemistry program package.Phys. Chem. Chem. Phys., 8:3172–3191, 2006.
[113] C.D. Sherrill, A.I. Krylov, E.F.C. Byrd, and M. Head-Gordon. Energies and ana-lytic gradients for a coupled-cluster doubles model using variational Bruecknerorbitals: Application to symmetry breaking in O+
4 . J. Chem. Phys., 109:4171,1998.
[114] M. Sierka and J. Sauer. Finding transition structures in extended systems: A strat-egy based on a combined quantum mechanics-empirical valence bond approach.J. Chem. Phys., 112(16):6983–6996, 2000.
[115] M.O. Sinnokrot and C.D. Sherrill. High-accuracy quantum mechanical studies ofπ − π interactions in benzene dimers.J. Phys. Chem. A, 110(37):10656–10668,2006.
[116] M.O. Sinnokrot, E.F. Valeev, and C.D. Sherrill. Estimates of the ab initio limit forπ−π interactions: The benzene dimer.J. Am. Chem. Soc., 124(36):10887–10893,2002.
[117] L.V. Slipchenko and M.S. Gordon. Electrostatic energy in the effective fragmentpotential method: Theory and application to benzene dimer.J. Comput. Chem.,28(1):276–291, 2007.
[118] L.V. Slipchenko and M.S. Gordon. Damping functions in the effective fragmentpotential method.Mol. Phys., 2009.
[119] L.V. Slipchenko and A.I. Krylov. Singlet-triplet gaps in diradicals by the spin-flipapproach: A benchmark study.J. Chem. Phys., 117:4694–4708, 2002.
[120] L.V. Slipchenko and A.I. Krylov. Efficient strategies for accurate calculationsof electronic excitation and ionization energies: Theory and application to thedehydro-meta-xylylene anion.J. Phys. Chem. A, 110:291–298, 2006.
113
[121] T. Smith, L.V. Slipchenko, and M.S. Gordon. Modelingπ − π interactions withthe effective fragment potential method: The benzene dimer and substituents.J.Phys. Chem. A, 112(23):5286–5294, 2008.
[122] D.B. Smithrud and F. Diederich. Strength of molecular complexation of apolarsolutes in water and in organic solvents is predictable by linear free energy rela-tionships: A general model for solvation effects on apolar binding.J. Am. Chem.Soc., 112(1):339–343, 1990.
[123] J.F. Stanton. Coupled-cluster theory, pseudo-Jahn-Teller effects and conical inter-sections.J. Chem. Phys., 115:10382–10393, 2001.
[124] J.F. Stanton, J. Gauss, J.D. Watts, W.J. Lauderdale, and R.J. Bartlett. ACES II,1993. The package also contains modified versions of the MOLECULE Gaussianintegral program of J. Almlof and P.R. Taylor, the ABACUS integral derivativeprogram written by T.U. Helgaker, H.J.Aa. Jensen, P. Jørgensen and P.R. Taylor,and the PROPS property evaluation integral code of P.R. Taylor.
[125] A.J. Stone. Distributed multipole analysis, or how to describe a molecular chargedistribution.Chem. Phys. Lett., 83(2):233 – 239, 1981.
[126] A.J. Stone. InThe Theory of Intermolecular Forces, pages 50 – 62. Oxford Uni-versity Press, 1997.
[127] A.J. Stone. Distributed multipole analysis: Stability for large basis sets.J. Chem.Theory Comput., 1(6):1128–1132, 2005.
[128] A.J. Stone and M. Alderton. Distributed multipole analysis - methods and appli-cations.Mol. Phys., 56(5):1047–1064, 1985.
[129] K.T. Tang and J.P. Toennies. An improved simple model for the van der Waalspotential based on universal damping functions for the dispersion coefficients.J.Chem. Phys., 80(8):3726–3741, 1984.
[130] T.P. Tauer and C.D. Sherrill. Beyond the benzene dimer: An investigation of theadditivity of π − π interactions.J. Phys. Chem. A, 109(46):10475–10478, 2005.
[131] S. Tsuzuki, T. Uchimaru, K. Matsumura, M. Mikami, and K. Tanabe. Effects ofthe higher electron correlation correction on the calculated intermolecular inter-action energies of benzene and naphthalene dimers: comparison between MP2and CCSD(T) calculations.Chem. Phys. Lett., 319(5-6):547 – 554, 2000.
[132] V. Vanovschi, A.I. Krylov, and P.G. Wenthold. Structure, vibrational frequen-cies, ionization energies, and photoelectron spectrum of the para-benzyne radicalanion.Theor. Chim. Acta, 120:45–58, 2008.
114
[133] H.H. Wenk, M. Winkler, and W. Sander. One century of aryne chemistry.Angew.Chem. Int. Ed. Engl., 42:502–528, 2003.
[134] P.G. Wenthold, J. Hu, and R.R. Squires.J. Am. Chem. Soc., 118:11865–11871,1996.
[135] P.G. Wenthold, R.R. Squires, and W.C. Lineberger. Ultraviolet photoelectronspectroscopy of the o-, m-, and p-benzyne negative ions. Electron affinities andsinglet-triplet splittings for o-, m-, and p-benzyne.J. Am. Chem. Soc., 120:5279–5290, 1998.
[136] C.A. White, B.G. Johnson, P.M.W. Gill, and M. Head-Gordon. Linear scalingdensity functional calculations via the continuous fast multipole method.Chem.Phys. Lett., 253(3-4):268–278, 1996.
[137] A.K. Wilson and T.H. Dunning. Benchmark calculations with correlated molecu-lar wave functions. 10. Comparison with ”exact” mp2 calculations on ne, hf, h2o,and n-2.J. Chem. Phys., 106(21):8718–8726, 1997.
[138] M. Wladyslawski and M. Nooijen. The photoelectron spectrum of the NO3 radi-cal revisited: A theoretical investigation of potential energy surfaces and conicalintersections. InACS Symposium Series, volume 828, pages 65–92, 2002.
[139] M. Wolfsberg and L. Helmholz. The spectra and electronic structure of the tetra-hedral ions MnO4-, CrO4–, and ClO4-.J. Chem. Phys., 20(5):837–843, 1952.
[140] D. E. Woon and T. H. Dunning Jr. Gaussian basis sets for use in correlatedmolecular calculations. iv. calculation of static electrical response properties.J.Chem. Phys., 100, 1994.
[141] D.E. Woon. Benchmark calculations with correlated molecular wave-functions.5. The determination of accurate ab-initio intermolecular potentials for he-2, ne-2, and ar-2.J. Chem. Phys., 100(4):2838–2850, 1994.
[142] D.E. Woon and T.H. Dunning. Benchmark calculations with correlated molecularwave-functions. 1. Multireference configuration-interaction calculations for the2nd-row diatomic hydrides.J. Chem. Phys., 99(3):1914–1929, 1993.
[143] D.E. Woon and T.H. Dunning. Benchmark calculations with correlated molecularwave-functions. 6. 2nd-row-A(2) and first-row 2nd-row-AB diatomic-molecules.J. Chem. Phys., 101(10):8877–8893, 1994.
[144] D.E. Woon, K.A. Peterson, and T.H. Dunning. Benchmark calculations withcorrelated molecular wave functions. IX. The weakly bound complexes Ar-H-2and Ar-HCl. J. Chem. Phys., 109(6):2233–2241, 1998.
115
[145] X. Ye, Z. Li, W. Wang, K. Fan, W. Xu, and Z. Hua. The parallelπ − π stacking:a model study with MP2 and DFT methods.Chem. Phys. Lett., 397(1-3):56 – 61,2004.
[146] S. Yoo, F. Zahariev, S. Sok, and M.S. Gordon. Solvent effects on optical proper-ties of molecules: A combined time-dependent density functional theory/effectivefragment potential approach.J. Chem. Phys., 129(14):8, 2008.
[147] T. Zeidan, M. Manoharan, and I.V. Alabugin. Ortho effect in the Bergmancyclization: Interception of p-benzyne intermediate by intramolecular hydrogenabstraction.J. Org. Chem., 71:954–961, 2006.
[148] Y. Zhang and W. Yang. A challenge for density functionals: Self-interaction errorincreases for systems with a noninteger number of electrons.J. Chem. Phys.,109(7):2604–2608, 1998.
116
Appendix A
GAMESS input for generating EFP
parameters of benzene
The GAMESS input file for generation the EFP parameters at the HF/6-311G++(3df,2p)
level of theory for the benzene monomer used in this work is given below.
$CONTRL RUNTYP=MAKEFP ISPHER=1 $END$SYSTEM TIMLIM=99999 MWORDS=150 $END$BASIS GBASIS=N311 NGAUSS=6 NFFUNC=1 NDFUNC=3 NPFUNC=2
DIFFSP=.T. DIFFS=.T. $END$SCF SOSCF=.F. DIIS=.T. $END$STONE
BIGEXP=4.0$END$DAMP IFTTYP(1)=3,2 IFTFIX(1)=1,1 THRSH=500 $END$DAMPGSC3=C1H4=H2C5=C1H6=H2C7=C1H8=H2C9=C1H10=H2C11=C1H12=H2BO43=BO21BO53=BO31BO65=BO21BO75=BO31BO87=BO21BO97=BO31BO109=BO21BO111=BO31BO119=BO31BO1211=BO21$END$DATA
benzene
117
C1C1 6.0 -1.3942300 0.00000000 0.0000000H2 1.0 -2.4764870 0.00000000 0.0000000C3 6.0 -0.6971150 0.00000000 1.2074388H4 1.0 -1.2382435 0.00000000 2.1447002C5 6.0 0.6971150 0.00000000 1.2074388H6 1.0 1.2382435 0.00000000 2.1447002C7 6.0 1.3942300 0.00000000 -0.0000000H8 1.0 2.4764870 0.00000000 0.0000000C9 6.0 0.6971150 0.00000000 -1.2074388H10 1.0 1.2382435 0.00000000 -2.1447002C11 6.0 -0.6971150 0.00000000 -1.2074388H12 1.0 -1.2382435 0.00000000 -2.1447002
$END
118
Appendix B
Geometries of benezene oligomers
Geometries of all the benzene oligomers studied in this work are provided below in the
format of the Q-Chem$efp fragmentssection. Fig. 1.16 provides the definition of the
format.
Sandwich (S) structure of benzene dimer:
benzene 0.0 0.0 0.0benzene 0.0 4.0 0.0
T-shaped (T) structure of benzene dimer:
benzene 0.0 0.0 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 5.2 0.0 1.57079633 1.57079633 0.00000000
Parallel-displaced (PD) structure of benzene dimer:
benzene 0.0 0.0 0.0benzene 1.2 3.9 0.0
S benzene trimer:
benzene 0.0 0.0 0.0benzene 0.0 4.0 0.0benzene 0.0 8.0 0.0
T1 benzene trimer:
benzene 0.0 0.0 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 5.2 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 10.4 0.0 1.57079633 1.57079633 1.57079633
T2 benzene trimer:
benzene 0.0 0.0 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 5.2 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 -5.2 0.0 1.57079633 1.57079633 0.00000000
PD benzene trimer:
119
benzene 0.0 0.0 0.0benzene 1.2 3.9 0.0benzene 0.0 7.8 0.0
C benzene trimer:
benzene 2.6 0.0 0.0 -0.25943951 0.00000000 0.00000000benzene -2.6 0.0 0.0 1.83495559 0.00000000 0.00000000benzene 0.0 -4.503332 0.0 3.92935069 0.00000000 0.00000000
PD-T benzene trimer:
benzene 0.0 0.0 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 5.2 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 -3.9 1.2 1.57079633 1.57079633 1.57079633
PD-S benzene trimer:
benzene 0.0 -4.0 0.0benzene 0.0 0.0 0.0benzene 1.2 3.9 0.0
S-T benzene trimer:
benzene 0.0 0.0 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 5.2 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 -4.0 0.0 1.57079633 1.57079633 1.57079633
S-10 benzene oligomer (larger S-oligomers are constructed similarly):
benzene 0.0 0.0 0.0benzene 0.0 4.0 0.0benzene 0.0 8.0 0.0benzene 0.0 12.0 0.0benzene 0.0 16.0 0.0benzene 0.0 20.0 0.0benzene 0.0 24.0 0.0benzene 0.0 28.0 0.0benzene 0.0 32.0 0.0benzene 0.0 36.0 0.0
T-10 benzene oligomer (larger T-oligomers are constructed similarly):
benzene 0.0 0.0 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 5.2 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 10.4 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 15.6 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 20.8 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 26.0 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 31.2 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 36.4 0.0 1.57079633 1.57079633 0.00000000benzene 0.0 41.6 0.0 1.57079633 1.57079633 1.57079633benzene 0.0 46.8 0.0 1.57079633 1.57079633 0.00000000
120
PD-10 benzene oligomer (larger PD-oligomers are constructed similarly):
benzene 0.0 0.0 0.0benzene 1.2 3.9 0.0benzene 0.0 7.8 0.0benzene 1.2 11.7 0.0benzene 0.0 15.6 0.0benzene 1.2 19.5 0.0benzene 0.0 23.4 0.0benzene 1.2 27.3 0.0benzene 0.0 31.2 0.0benzene 1.2 35.1 0.0
121
Appendix C
Converter of the EFP parameters
A script, which converts the GAMESS input parameters to the Q-Chem format is given
below:
#!/usr/bin/env python"""EFP parameters converter.Converts from GAMESS to Q-Chem input format.
Usage: python converter.py <GAMESS INPUT>"""
import sysimport re
# Bohrs to Angstroms convertion factorB2A = 0.529177249
__version__ = "1.15"__author__ = "Vitalii Vanovschi"
if "set" not in globals():from sets import Set as set
(FRAG, COORDS, MONOPOLES, DIPOLES, QUADRUPOLES, OCTUPOLES, \BASIS, WAVEFUNCTION, FOCK, LMO, MULTIPLICITY, SCREEN2,\POLARIZABLE, DYNAMIC, CANONVEC, CANONFOK, SCREEN3) = xrange(0, 17)
def dump_qchem_frag(frag_name, labels, charges, dipoles, quadrupoles, \octupoles, wavefunction, fock, centroids, screen2, \ordered_labels, nuc_charges, pols, disps, basis,coordinates):
"""Outputs EFP fragment parameters in the Q-Chem format"""
print "fragment", frag_name
# output atomsfor coordinate in coordinates:
if coordinate[0][0:2] != "bo":print coordinate[0].rstrip("0123456789") + " " \
+ " ".join(coordinate[1:])
# output atomic charges (as multipoles)for label in ordered_labels:
coord = labels[label]if label in nuc_charges and float(nuc_charges[label]) > 0.1:
print "mult ", " ".join(coord)print nuc_charges[label]
# output multipolesfor label in ordered_labels:
122
coord = labels[label]print "mult ", " ".join(coord)if label in screen2:
print "cdamp " + screen2[label][1]if label in charges:
print charges[label]if label in dipoles:
print " ".join(dipoles[label])if label in quadrupoles:
print " ".join(quadrupoles[label])if label in octupoles:
print " ".join(octupoles[label])
# output polarization pointsfor pol in pols:
print "pol ", pol[1], " ", pol[2], " ", pol[3]print " ".join(pol[4:])
# output dispersion pointsdispindexes = list(set([x[0] for x in disps]))dispindexes.sort()for i in dispindexes:
isfirst = Truefor disp in disps:
if disp[0] == i:if isfirst:
print "disp ", disp[1], " ", disp[2], " ", disp[3]isfirst = False
print disp[4],print
# output exchange-repulsionprint "er_basis"print "er_wavefunction"wfindexes = wavefunction.keys()wfindexes.sort()
if len(wfindexes) * (len(wfindexes) + 1) != 2 * len(fock):raise AssertionError("Inconsistent input.\n"
"The following assertion has failed:\n""len(wfindexes) * (len(wfindexes) + 1) == 2 * len(fock)\n""Values: len(wfindexes) = " + str(len(wfindexes)) + ", ""len(fock) = " + str(len(fock)) + ".\n""Did you remove ’>’ from the last line of the FOCK MATRIX \n""section in GAMESS input? (an old GAMESS bug)")
for ind in wfindexes:orb = wavefunction[ind]i = 0for shell in basis:
if shell == ’s’:i += 1
if shell == ’p’:i += 3
if shell == ’sp’ or shell == ’l’:i += 4
if shell == ’d’:tmp = orb[i:i+6]
#10 -> 10#11 -> 12#12 -> 15#13 -> 11#14 -> 13
123
#15 -> 14orb[i] = tmp[0]orb[i+1] = tmp[3]orb[i+2] = tmp[1]orb[i+3] = tmp[4]orb[i+4] = tmp[5]orb[i+5] = tmp[2]i += 6
if shell == ’f’:tmp = orb[i:i+10]orb[i] = tmp[0] #fxxxorb[i+1] = tmp[3] #fxxyorb[i+2] = tmp[5] #fxyyorb[i+3] = tmp[1] #fyyyorb[i+4] = tmp[4] #fxxzorb[i+5] = tmp[9] #fxyzorb[i+6] = tmp[6] #fyyzorb[i+7] = tmp[7] #fxzzorb[i+8] = tmp[8] #fyzzorb[i+9] = tmp[2] #fzzzi += 10
for i in range(len(orb)):if i % 5 == 0:
print ind,if orb[i][0] == "-":
print orb[i],else:
print " " + orb[i],if i % 5 == 4:
printprint
print "er_fock_matrix"for i in range(len(fock)):
print fock[i],if i % 5 == 4:
printprintprint "er_lmos"for lmo in centroids:
print " ".join(lmo)
def parse_gamess(gamess_file):"""Parses GAMESS input file"""
str2int = "frag" : FRAG,"coordinates" : COORDS,"monopoles" : MONOPOLES,"dipoles" : DIPOLES,"quadrupoles" : QUADRUPOLES,"octupoles" : OCTUPOLES,"basis" : BASIS,"wavefunction" : WAVEFUNCTION,"fock" : FOCK,"lmo" : LMO,"multiplicity" : MULTIPLICITY,"screen2": SCREEN2,"polarizable": POLARIZABLE,"dynamic": DYNAMIC,"canonvec": CANONVEC,"canonfok": CANONFOK,
124
"screen3": SCREEN3
frag_name = Noneordered_labels = []labels = coordinates = []charges = nuc_charges = dipoles = quadrupoles = octupoles = screen2 = centroids = []wavefunction = fock = []expect = FRAGprefix = ""pols = []disps = []basis = []
print "$efp_params"for line1 in gamess_file:
line_orig = line1line1 = line1.strip()
if not line1 or line1.lower() == ’STOP’:continue
if line1[-1] == ">":prefix += " " + line1continue
else:line = prefix + " " + line1prefix = ""
fragment_start = re.search("\$([A-Za-z0-9]+)", line)if fragment_start:
text = fragment_start.groups()[0].lower()if text in ("efrag", "contrl"):
continue
if text == "end" and frag_name:# print "frag_name:", frag_name# print "labels:", labels# print "dipoles:", dipoles# print "quadrupoles:", quadrupoles# print "octupoles:", octupoles# print "wavefunction:", wavefunction# print "fock:", fock# print "centroids:", centroids
dump_qchem_frag(frag_name, labels, charges, dipoles,quadrupoles, octupoles, wavefunction, fock, centroids,screen2, ordered_labels, nuc_charges, pols, disps,basis, coordinates)
frag_name = Nonelabels = charges = nuc_charges = dipoles =
125
quadrupoles = octupoles = screen2 = centroids = []coordinates = []ordered_labels = []wavefunction = fock = []expect = FRAGprefix = ""pols = []disps = []basis = []
if text != "end":frag_name = textwhile 1:
line = gamess_file.next()words = re.findall("([A-Za-z0-9\.+-]+)", line)if words[0].lower() == "coordinates":
breakexpect = COORDS
continue
words = re.findall("([A-Za-z0-9\.+-]+)", line)if words:
firstword = words[0].lower()else:
firstword = None
if firstword in str2int:expect = str2int[firstword]continue
if firstword == ’projection’ and len(words) > 1:expect = str2int[words[1].lower()]continue
if expect == COORDS and len(words) > 3:atom_x = str(float(words[1])*B2A)atom_y = str(float(words[2])*B2A)atom_z = str(float(words[3])*B2A)labels[firstword] = (atom_x, atom_y, atom_z)coordinates.append((firstword, atom_x, atom_y, atom_z))ordered_labels.append(firstword)
if expect == MONOPOLES and len(words) > 1:charges[firstword] = words[1]nuc_charges[firstword] = words[2]
if expect == DIPOLES and len(words) == 4:dipoles[firstword] = words[1:]
if expect == QUADRUPOLES and len(words) == 7:quadrupoles[firstword] = words[1:]
if expect == OCTUPOLES and len(words) == 11:octupoles[firstword] = words[1:]
if expect == WAVEFUNCTION:# 3 1-1.40747045E-01 2.10323654E-01 2.42989533E-01-4.68907985E-01-1.03447318E-01
key = int(line_orig[:3].strip())
126
if key not in wavefunction:wavefunction[key] = []
values = line_orig[5:].rstrip()while values:
wavefunction[key].append(values[:15].strip())values = values[15:]
if expect == FOCK:fock += words
if expect == LMO and len(words) == 4:centroids.append([str(float(x)*B2A) for x in words[1:]])
if expect == SCREEN2:screen2[firstword] = words[1:]
if expect == POLARIZABLE:if len(words) == 4:
pol_x = str(float(words[1])*B2A)pol_y = str(float(words[2])*B2A)pol_z = str(float(words[3])*B2A)pols.append([words[0], pol_x, pol_y, pol_z])
elif len(words) == 9:pols[-1].extend(words)
if expect == DYNAMIC:if len(words) == 5 or len(words) > 5 and words[5] == "--":
index = int(words[1])disp_x = str(float(words[2])*B2A)disp_y = str(float(words[3])*B2A)disp_z = str(float(words[4])*B2A)disps.append([index, disp_x, disp_y, disp_z])
elif len(words) == 9:disps[-1].append(sum([float(x) for x in words[:3]]) / 3)
if expect == BASIS:if firstword in (’s’, ’sp’, ’p’, ’d’, ’l’, ’f’):
basis.append(firstword)
print "$end"
if __name__ == "__main__":if len(sys.argv) != 2:
print "Usage: python converter.py <GAMESS INPUT>"sys.exit()
INPUT = open(sys.argv[1])parse_gamess(INPUT)INPUT.close()
127
Appendix D
Fragment-fragment electrostatic
interaction
The equations for electrostatics inter-fragment interaction energy are presented below.
k and l are the multipole expansion points.q is the charge,µ is the dipole,Θ is the
quadrupole,Ω is the octupole,R is the distance between the expansion pointsk andl,
a, b andc are the components of the distance.
Charge-charge interaction energy:
Ech−chkl =
qkql
Rkl
(D.1)
Charge-dipole interaction energy:
Ech−dipkl =
qk∑x,y,z
a µlaa
R3kl
(D.2)
Charge-quadrupole interaction energy:
Ech−quadkl =
qk∑x,y,z
a
∑x,y,zb Θl
abab
R5kl
(D.3)
Charge-octupole interaction energy:
Ech−octkl =
qk∑x,y,z
a
∑x,y,zb
∑x,y,zc Ωl
abcabc
R7kl
(D.4)
128
Dipole-dipole interaction energy:
Edip−dipkl =
∑x,y,za µk
aµlb
R3kl
− 3
∑x,y,za
∑x,y,zb µk
aµlbab
R5kl
(D.5)
Dipole-quadrupole interaction energy:
Edip−quadkl = −2
∑x,y,za
∑x,y,zb Θk
abµlab
R5kl
+ 5
∑x,y,za
∑x,y,zb Θk
abab∑x,y,z
c µlcc
R7kl
(D.6)
Quadrupole-quadrupole interaction energy:
Equad−quadkl = 2
∑x,y,za
∑x,y,zb Θk
abΘlab
3R5kl
−20
∑x,y,za
∑x,y,zb Θk
abb∑x,y,z
c Θlacc
3R7kl
+ 35
∑x,y,za
∑x,y,zb Θk
abab
3R9kl
(D.7)
Nuclei-charge interaction energy:
Enuc−chIl =
ZIql
rkl
(D.8)
Nuclei-dipole interaction energy:
Enuc−dipIl =
ZI∑x,y,z
a µlaa
R3kl
(D.9)
Nuclei-quadrupole interaction energy:
Enuc−quadIl =
ZI∑x,y,z
a
∑x,y,zb Θl
abab
R5kl
(D.10)
Nuclei-octupole interaction energy:
Enuc−octIl =
ZI∑x,y,z
a
∑x,y,zb
∑x,y,zc Ωl
abcabc
R7kl
(D.11)
129
Appendix E
Buckingham multipoles
E.1 The conversion from the sperical to the Bucking-
ham multipoles
The Buckingham chargeq in terms of the spherical multipoleQ with l = 0 angular
momentum:
q = Q0,0 (E.1)
The Buckingham dipoleµ in terms of the spherical multipoleQ with l = 1 angular
momentum:
µx = −2Q11
µy = 2Q11
µz = Q10
(E.2)
130
The Buckingham quadrupoleΘ in terms of the spherical multipoleQ with l = 2
angular momentum:
Θxx = 6Q22 −Q20
Θxy = −6Q22
Θyy = −6Q22 −Q20
Θxz = −3Q21
Θyz = 3Q21
Θzz = 2Q20
(E.3)
The Buckingham octupoleΩ in terms of the spherical multipoleQ with l = 3 angular
momentum:
Ωxxx = −30Q33 + 6Q31
Ωxxy = 30Q33 − 2Q31
Ωxxz = −3Q30 + 10Q32
Ωxyy = 30Q33 + 6Q31
Ωxyz = −10Q32
Ωxzz = −8Q31
Ωyyy = −30Q33 − 6Q31
Ωyyz = −3Q30 − 10Q32
Ωyzz = 8Q31
Ωzzz = 6Q30
(E.4)
131
E.2 The conversion from the Cartesian to the Bucking-
ham multipoles
.
The Buckingham chargeq in terms of the Cartesian chargeq′:
q = q′ (E.5)
The Buckingham dipoleµ in terms of the Cartesian dipoleµ′:
µx = µ′x
µy = µ′y
µz = µ′z
(E.6)
The Buckingham quadrupoleΘ in terms of the Cartesian quadrupoleΘ′:
halftrace =Θ′
xx + Θ′yy + Θ′
zz
2
Θxx =3
2Θ′
xx − halftrace
Θxy =3
2Θ′
xy
Θyy =3
2Θ′
yy − halftrace
Θxz =3
2Θ′
xz
Θyz =3
2Θ′
yz
Θzz =3
2Θ′
zz − halftrace
(E.7)
132
The Buckingham octupoleΩ in terms of the Cartesian octupoleΩ′:
traceX = Ω′xxx + Ω′
xyy + Ω′xzz
traceY = Ω′xxy + Ω′
yyy + Ω′yzz
traceZ = Ω′xxz + Ω′
yyz + Ω′zzz
Ωxxx =5
2Ω′
xxx −3
2traceX
Ωxxy =5
2Ω′
xxy −3
2traceY
Ωxxz =5
2Ω′
xxz −3
2traceZ
Ωxyy =5
2Ω′
xyy −1
2traceY
Ωxyz =5
2Ω′
xyz −1
2traceZ
Ωxzz =5
2Ω′
xzz −1
2traceX
Ωyyy =5
2Ω′
yyy −1
2traceX
Ωyyz =5
2Ω′
yyz −1
2traceZ
Ωyzz =5
2Ω′
yzz −1
2traceY
Ωzzz =5
2Ω′
zzz
(E.8)
133
Appendix F
The Q-Chem input for EFP water
cluster
The Q-Chem input filea for a molecular system with two effective fragment water
molecules and another water moleucle in the the active region is presented below:
$molecule0 1O 2.98679899864 0.000000000000 0.000000000000H 3.37280799847 0.408736299796 0.759199899653H 3.37280799847 0.408736299796 -0.759199899653$end
$rembasis 6-31G*exchange hf$end
$efp_fragmentswater 2.5 0.0 0.0water 0.0 2.5 0.0$end
$efp_paramsfragment watero 2.98679899864 0.000000000000 0.000000000000h 3.37280799847 0.408736299796 0.759199899653h 3.37280799847 0.408736299796 -0.759199899653
mult 2.98679899864 0.000000000000 0.0000000000008.00000
mult 3.37280799847 0.408736299796 0.7591998996531.00000
mult 3.37280799847 0.408736299796 -0.759199899653
aLines ending at a backslash symbol should be joined with the next line in the actual Q-Chem input.
134
1.00000
mult 2.98679899864 0.000000000000 0.000000000000cdamp 2.432093890-5.9880774156
0.9069997308 0.960401737 0.000000000-2.0776300674 -2.055053110 -1.094067772 0.1972105926\
0.0000000000 0.0000000001.2431711201 1.307173512 0.0000000000 0.3882335020\0.0000000000 0.357964785 0.0000000000 0.3997408222\0.4232766191 0.000000000
mult 3.37280799847 0.408736299796 0.759199899653cdamp 1.856169014-0.6161272254-0.0607832521 -0.0643620268 -0.1385305992-0.2984353445 -0.2970178207 -0.2629326306 0.0123821246\
0.0186582649 0.0197568195-0.1419121249 -0.1496397602 -0.2838114377 -0.0466364802\-0.1010324765 -0.0434504063 -0.0996406903 -0.0238456713\-0.0252496482 0.0121573058
mult 3.37280799847 0.408736299796 -0.759199899653cdamp 1.856168614-0.6161272254-0.0607832521 -0.0643620268 0.1385305992-0.2984353445 -0.2970178207 -0.2629326306 0.0123821246\-0.0186582649 -0.0197568195-0.1419121249 -0.1496397602 0.2838114377 -0.0466364802\
0.1010324765 -0.0434504063 0.0996406903 -0.0238456713\-0.0252496482 -0.0121573058
mult 3.17980349856 0.204368149898 0.379599949853cdamp 10.000000000-1.3898340671
0.0671975530 0.0711539868 -0.0800483392-0.7200126389 -0.7080787920 -0.3459299814 0.1042426037\
0.1265126185 0.13396138360.0823824078 0.0860368352 -0.1633636547 0.0224998491\
-0.0684206741 0.0201192189 -0.0686914603 0.0517621518\0.0548097854 -0.0023653270
mult 3.17980349856 0.204368149898 -0.379599949853cdamp 10.000000000-1.3898340671
0.0671975530 0.0711539868 0.0800483392-0.7200126389 -0.7080787926 -0.3459299814 0.1042426037\-0.1265126185 -0.1339613836
0.0823824078 0.0860368352 0.1633636547 0.0224998491\0.0684206741 0.0201192189 0.0686914603 0.0517621518\
135
0.0548097854 0.0023653270
pol 3.225043346880 0.252271132154 -0.4488098121240.3364082841 0.3771885556 2.0857377185 0.3562152772\
-0.6146554330 -0.6508454378 0.3562153645 -0.9006532748\-0.9536819287
pol 3.22504358327 0.252271266354 0.4488097842360.3364078945 0.3771881750 2.0857352855 0.3562148807\0.6146555083 0.6508456481 0.3562149890 0.9006533941\0.9536821303
pol 2.68786044569 0.0680155240902 0.0000000000000.0546318425 0.2156829505 0.4678560525 0.0987167685\
-0.0000001586 -0.0000002666 0.1193632771 -0.0000002037\-0.0000002725
pol 3.07178559579 -0.294563387213 0.0000000000000.2027164107 0.0676015168 0.4678627585 0.1278414717\0.0000000833 0.0000000562 0.1071947676 0.0000000844\0.0000000709
disp 3.22504334688 0.252271132154 -0.4488098121240.9330994836 0.9328329368 0.9312726852 0.9258111592\0.9104538267 0.8718871966 0.7834519894 0.6086537072\0.3529710151 0.1263476070 0.0217816395 0.0007641188
disp 3.22504358327 0.252271266354 0.4488097842360.9330984158 0.9328318692 0.931271618733 0.92581009\0.9104527744 0.8718861726 0.7834510398 0.6086529349\0.3529705524 0.1263474433 0.0217816120 0.0007641179
disp 2.68786044569 0.068015524090 0.0000000000000.2460531659 0.2459405339 0.2452820477 0.2429879390\0.2366256365 0.2211913563 0.1883852733 0.1320296167\0.0651263370 0.0192960190 0.0029443470 0.0001007762
disp 3.07178559579 -0.294563387213 0.0000000000000.2460564460 0.2459438126 0.2452853183 0.2429911814\0.2366288005 0.2211943284 0.1883878314 0.1320314467\0.0651272751 0.0192963144 0.0029443942 0.0001007779
er_basis sto-3g
er_wavefunction1 9.12366435E-02 -2.55851988E-01 -2.74993695E-01 -2.91183082E-011 4.26891343E-01 1.09661524E-01 -5.16313349E-012 9.12367358E-02 -2.55852205E-01 -2.74994306E-01 -2.91183344E-012 -4.26890773E-01 -5.16313220E-01 1.09660818E-01
136
3 1.54680314E-01 -6.48520365E-01 7.87066922E-01 -1.96453849E-013 -1.20289533E-07 8.88260577E-02 8.88262341E-024 1.54680523E-01 -6.48521682E-01 -2.41107828E-01 7.74554226E-014 3.89668406E-08 8.88271814E-02 8.88271243E-02
er_fock_matrix-0.8097112963 -0.1811928224 -0.8097117054 -0.1951745608-0.1951748560 -0.5598738582 -0.1951742273 -0.1951744336-0.1696642207 -0.5598736581
er_lmos3.22504334688 0.252271132154 -0.4488098121243.22504358327 0.252271266354 0.4488097842362.68786044569 0.068015524090 0.0000000000003.07178559579 -0.294563387213 0.000000000000$end
137
Appendix G
Q-Chem’s keywords affecting the EFP
parameters computation
Q-Chem’s keywords affecting the computation of the electrostatics and polarization
parameters are presented below in the format compartiblie with the Q-Chem manual:
POL GENWhether distributed polarizabilities should be computed for makeefp jobtype.
TYPE:LOGICAL
DEFAULT:TRUE
OPTIONS:TRUE/FALSE
RECOMMENDATION:none
POL PRINT BOHRPrint polarizabilities in Bohr instead of Angstroms.
TYPE:LOGICAL
DEFAULT:FALSE
OPTIONS:TRUE/FALSE
RECOMMENDATION:none
138
POL DIFF TYPEFinite differentitation type
TYPE:INTEGER
DEFAULT:2
OPTIONS:2 two steps finite difference (for higher accuracy)
1 one step finite difference (faster)RECOMMENDATION:
none
POL FIELD DELTAElectric filed (step size for numerical differentiation of polarizabilities)
TYPE:INTEGER
DEFAULT:100
OPTIONS:n n ∗ 10−7 step size.
RECOMMENDATION:none
POL LMO FREEZEHow many core orbitals to freeze for LMO computation
TYPE:INTEGER
DEFAULT:-1
OPTIONS:-1 freeze all core orbitals
n freezen core orbitalsRECOMMENDATION:
none
DMA GENWhether to perform distributed multipole analysis (DMA) for jobtype makeefp.
TYPE:LOGICAL
DEFAULT:TRUE
OPTIONS:TRUE/FALSE
RECOMMENDATION:none
139
DMA MIDPOINTSAdd bond midpoint to DMA expansion points on bond centers (in addtion to expansionpoints on nucleai)
TYPE:LOGICAL
DEFAULT:TRUE
OPTIONS:TRUE/FALSE
RECOMMENDATION:none
DAMP STO1G GUESSWhether to use STO1G guess procedure
TYPE:LOGICAL
DEFAULT:TRUE
OPTIONS:TRUE/FALSE
RECOMMENDATION:none
DAMP GEN EXP FITWhether to compute exponential damping
TYPE:LOGICAL
DEFAULT:FALSE
OPTIONS:TRUE/FALSE
RECOMMENDATION:none
DAMP GEN GAUSS FITWhether to compute gaussian damping
TYPE:LOGICAL
DEFAULT:FALSE
OPTIONS:TRUE/FALSE
RECOMMENDATION:none
140
DAMP EXP TOLTolerance for exponential fitting
TYPE:INTEGER
DEFAULT:10
OPTIONS:n n ∗ 10−7 tolerance
RECOMMENDATION:none
DAMP GAUSS TOLTolerance for gaussian fitting
TYPE:INTEGER
DEFAULT:1000
OPTIONS:n n ∗ 10−7 tolerance
RECOMMENDATION:none
DAMP GRID STEPGrid step for damping of electrostatic potential
TYPE:INTEGER
DEFAULT:50
OPTIONS:n n/100 Angstroms
RECOMMENDATION:none
DAMP RMINMinimum distance from a nucleai to a grid point
TYPE:INTEGER
DEFAULT:67
OPTIONS:n n/100 Angstroms
RECOMMENDATION:none
141
DAMP RMAXMaximum distance from a nucleai to a grid point
TYPE:INTEGER
DEFAULT:300
OPTIONS:n n/100 Angstroms
RECOMMENDATION:none
142