Overview of Molecular Modelling and

Ab initio Molecular Orbital Methods

Suitable for Use with Energetic Materials

H. Dorsett and A. White

Weapons Systems Division

Aeronautical and Maritime Research Laboratory

DSTO-GD-0253

ABSTRACT

This is a review of molecular modelling techniques which may be applied to studies of

energetic materials. It focusses on ab initio ('first-principles') molecular orbital
calculations, since these methods offer the greatest accuracy. Since ab initio calculations
are very computer-intensive, approximate MO methods are also discussed, which offer
reasonably accurate predictions with reduced calculation times. These approximate
methods include density functional theory and 'layered' techniques which combine
different levels of theoretical sophistication into one calculation.

RELEASE LIMITATION

Approved for public release

Published by

DSTO Aeronautical and Maritime Research Laboratory

PO Box 1500
Salisbury South Australia 5108 Australia

Telephone: (08) 8259 5555

Fax: (08) 8259 6567
© Commonwealth of Australia 2000
AR-011-578
September 2000

APPROVED FOR PUBLIC RELEASE

 Overview of Molecular Modelling and
Ab initio Molecular Orbital Methods
Suitable for Use with Energetic Materials

Executive Summary
Qualifying energetic materials (EM) and formulations for military use is a tedious,
expensive and hazardous process. It may be possible to replace much of the redundant
empirical testing of EM, either in service or under consideration, by computer
simulation. Therefore, accurate theoretical models of EM performance and stability can

• Accelerate research and development of new materials with improved

performance and handling characteristics
• Reduce maintenance costs for energetic materials in service and
• Reduce risk to personnel testing, handling or using energetic materials.

Many relevant characteristics of energetic materials - for instance, the performance and
sensitivity of explosives, or the stability and signatures of propellants – are controlled
by their chemical properties. These characteristics depend upon properties such as
molecular structures and reaction energies that can be predicted with computational
chemistry techniques which focus on molecular modelling.

All molecular modelling techniques can be classified under three general categories: ab
initio electronic structure calculations, semi-empirical methods, and molecular
mechanics. Ab initio or ‘first principles’ electronic structure methods are based upon
quantum mechanics and therefore provide the most accurate and consistent
predictions for chemical systems. However ab initio methods are extremely computer-
intensive. Semi-empirical methods are also founded upon quantum mechanics, but
speed computation by replacing some explicit calculations with approximations based
upon experimental data. Molecular mechanics techniques are purely empirical
methods based upon the principles of classical physics, and as such are
computationally fast. Molecular mechanics methods completely neglect explicit
treatment of electronic structure, and are therefore severely limited in scope; however,
they often provide the only means with which to study very large chemical systems
(e.g., polymers or solutions) or non-homogeneous mixtures like those typically used in
conventional explosives, pyrotechnics or propellants formulations.

Ab initio methods are capable of high accuracy predictions over a wide range of
systems. Rapid advances in computer technology are making ab initio methods
increasingly more practical for use with realistic chemical systems. Similarly,
computationally ‘cheaper’ techniques such as density functional calculations and
‘layered’ methods such as ONIOM, are continually being refined. These methods show
promise of providing consistent and accurate chemical predictions for complicated
systems requiring explicit treatment of electronic structure, such as energetic molecules
that contain azides or nitrogen oxides.

This document describes the theoretical basis of molecular modelling techniques which
may be applied to energetic materials. A subsequent publication will describe in detail
the application of electronic structure methods to the study of energetic materials.
 Contents

1. INTRODUCTION.............................................................................................................. 1

2. SURVEY OF COMPUTATIONAL CHEMISTRY METHODS ................................. 2

2.1 Ab initio electronic structure methods............................................................................ 2
2.2 Semi-empirical methods.................................................................................................... 4
2.2.1 The extended Hückel method ................................................................................ 4
2.2.2 Neglect of differential overlap (NDO) .................................................................. 4
2.2.3 Neglect of diatomic differential overlap (NDDO)............................................... 5
2.3 Molecular mechanics.......................................................................................................... 5
2.4 Hybrid methods .................................................................................................................. 7

3. AB INITIO MOLECULAR ORBITAL THEORY .......................................................... 8

3.1 The physico-chemical model............................................................................................ 8
3.1.1 The molecular Hamiltonian.................................................................................... 9
3.2 Hartree-Fock theory............................................................................................................ 9
3.2.1 The Schrödinger equation....................................................................................... 9
3.2.2 Antisymmetry and electron spin ......................................................................... 10
3.2.3 Ab initio essentials .................................................................................................. 11
3.2.4 Approximations ..................................................................................................... 12
3.2.4.1 Non-relativistic solutions ............................................................................... 12
3.2.4.2 Born-Oppenheimer approximation............................................................... 12
3.2.4.3 Single particle approximation........................................................................ 12
3.2.4.4 Linear combinations of atomic orbitals (LCAO) ......................................... 12
3.2.5 The variational principle and Roothan-Hall equations. ................................... 14
3.3 Basis Sets ............................................................................................................................ 14
3.3.1 Split valence basis sets........................................................................................... 15
3.3.2 Polarised basis sets................................................................................................. 16
3.3.3 Diffuse basis sets .................................................................................................... 16
3.3.4 High angular momentum basis sets.................................................................... 16
3.4 Electron correlation .......................................................................................................... 17
3.4.1 Configuration-interaction ..................................................................................... 17
3.4.2 Møller-Plesset perturbation .................................................................................. 18
3.4.3 Density-functional theory ..................................................................................... 19
3.4.4 Caveats for post-HF methods............................................................................... 20
3.5 Open-shell calculations and spin contamination....................................................... 21

4. ACCURACY VERSUS EFFICIENCY FOR AB INITIO METHODS ....................... 23

4.1 Error reduction .................................................................................................................. 23
4.1.1 Methods................................................................................................................... 23
4.1.2 Basis sets.................................................................................................................. 23
4.2 Resource usage .................................................................................................................. 24
4.3 High-accuracy/low-cost model chemistries ................................................................. 25
4.3.1 Compound methods.............................................................................................. 25
4.3.2 Complete basis set methods ................................................................................. 26
4.3.3 Caveats for energetic materials ............................................................................ 27

5. CONCLUSIONS .............................................................................................................. 27
6. REFERENCES ................................................................................................................... 29
 DSTO-GD-0253

1. Introduction
Computational chemistry may be defined as the application of mathematical and
theoretical principles to the solution of chemical problems. Molecular modelling, a
subset of computational chemistry, concentrates on predicting the behaviour of
individual molecules within a chemical system. The most accurate molecular models
use ab initio or ‘first principles’ electronic structure methods, based upon the principles
of quantum mechanics, and are generally very computer-intensive. However, due to
advances in computer storage capacity and processor performance, molecular
modelling has been a rapidly evolving and expanding field, to the point that it is now
possible to solve relevant problems in an acceptable amount of time.

The types of predictions possible for molecules and reactions include [1]:

• Heats of formation
• Bond and reaction energies
• Molecular energies and structures (thermochemical stability)
• Energies and structures of transition states (activation energies)
• Reaction pathways, kinetics and mechanisms
• Charge distribution in molecules (reactive sites)
• Substituent effects
• Electron affinities and ionisation potentials
• Vibrational frequencies (IR and Raman spectra)
• Electronic transitions (UV/Visible spectra)
• Magnetic shielding effects (NMR spectra)

Prediction of these properties has many applications in energetic materials research,

including studies of synthesis pathways, reaction products, initiation mechanisms, and
exhaust plume signature. This report presents a review of computational chemistry
techniques, focussing on electronic structure methods. Electronic structure methods,
particularly ab initio calculations, are capable of consistent predictions with high
accuracy (± 20 kJ/mol)1 over a wide range of systems – a critical prerequisite for the
successful modelling of energetic materials.

An example of the application of molecular modelling to energetic materials is

calculation of the detonation properties of a new explosive FOX-7 (1,1-diamino-2,2-
dinitroethylene). The ideal detonation properties of any explosive formulation can be
calculated from the principles of chemical thermodynamics using only the heats of
formation of the ingredients as input. The experimental heat of formation of FOX-7 is

1 The unit ‘kJ/mol’ is a convenient scale for discussing molecular energies. For instance, bonds
within molecules typically have energies of a few hundred kJ/mol; bonds between molecules
(e.g., hydrogen bonds) have energies of tens of kJ/mol; and bonds between atoms of inert gases
(e.g., He, Ne, Ar) have energies no more than a few kJ/mol.

1
DSTO-GD-0253

-134 kJ/mol [2], and the calculated heat of formation using the PM-3 semi-empirical
method is –53 kJ/mol. Estimates of the velocity of detonation (VoD) and Chapman-
Jouguet detonation pressure (PC-J) for FOX-7 calculated with the CHEETAH
thermochemical code [3] are presented in Table 1.1. In this case, it is clear that the PM-3
heat of formation is accurate enough to yield values of VoD and PC-J that are very close
to those estimated from the experimental heat of formation. Heats of formation
calculated with ab initio methods are nearly identical to the experimental value.

Table 1.1 FOX-7 detonation parameters estimated with CHEETAH.

∆Hf CHEETAH v(1.39) CHEETAH v1.41
VoD PC-J VoD PC-J
-53 kJ/mol 9130 m/s 37.1 GPa 9040 m/s 36.0 GPa
-134 kJ/mol 9090 m/s 36.6 GPa 8870 m/s 34.0 GPa

Ab initio, semi-empirical and molecular mechanics methods have been used extensively
to study the chemistry of energetic materials in explosives, propellants and
pyrotechnics. A comprehensive review of published studies is beyond the scope of this
paper, but will be summarised in a subsequent technical report [4].

2. Survey of Computational Chemistry Methods

All molecular modelling techniques can be classified under three major categories:
ab initio electronic structure calculations, semi-empirical methods and molecular
mechanics. General characteristics for each method are summarised in Table 2.1.

2.1 Ab initio electronic structure methods

Of the three, ab initio molecular orbital methods are the most accurate and consistent
because they provide the best mathematical approximation to the actual system. The
term ab initio implies that the computations are based solely on the laws of quantum
mechanics, the masses and charges of electrons and atomic nuclei, and the values of
fundamental physical constants, such as the speed of light (c = 2.998×108 m/s) or
Planck’s constant (h = 6.626×10-34 J·s), and contain no approximations. ‘Molecular
orbital’ methods solve Schrödinger’s equation for the chemical system using a ‘basis
set’ of functions that satisfy a series of rigorous mathematical approximations. Ab initio
concepts are outlined in Chapter 3.

Ab initio molecular orbital calculations are specified by a ‘model chemistry.’ The model
chemistry includes the choice of method and basis set, the general structure and

2
 DSTO-GD-0253

electronic state of the molecular system under study (e.g., charge and spin states), and
the treatment of electron spin2. Molecular properties can be assessed from a user-
specified input (single-point energy calculation or SPE), or the molecule can be allowed
to relax to a minimum energy configuration (geometry optimisation).

Table 2.1. Synopsis of molecular modelling techniques.

Method Advantages Disadvantages Best for
Ab initio • Useful for a • Computationally • Small systems
• Uses quantum broad range of expensive (tens of atoms)
physics systems • Electronic
• Mathematically • Does not depend transitions
rigorous: no on experimental • Systems without
empirical data experimental
parameters • Calculates data
transition states • Systems
and excited states requiring high
accuracy.

Semi-empirical
• Uses quantum • Less demanding • Requires ab initio • Medium-sized
physics computationally or experimental systems
• Uses than ab initio data for (hundreds of
experimental methods. parameters. atoms).
parameters • Calculates • Less rigorous than • Electronic
• Uses extensive transition states ab initio methods. transitions.
approximations and excited
states.

Molecular
Mechanics
• Uses classical • Computationally • Does not calculate • Large systems
physics ‘cheap:’ fast and electronic (thousands of
• Relies on force useful with properties atoms)
field with limited computer • Requires ab initio • Systems or
embedded resources. or experimental processes that do
empirical • Can be used for data for not involve bond
parameters. large molecules parameters breaking.
like enzymes. • Commercial
software
applicable to a
limited range of
molecules

Ab initio molecular orbital computations can provide accurate quantitative predictions

of chemical properties for a wide range of molecular systems. However, they place a
considerable demand on computer resources. The choice of theoretical method and
basis set determine the duration of the calculation; thus, a sophisticated method and a

2Electrons may be assigned singly (unrestricted) or in pairs of opposite spin (restricted) to the
molecular orbitals that make up the total wavefunction of the system.

3
DSTO-GD-0253

large basis set will provide more accurate results, but will also require more computer
resources. Hence timely calculations on large molecules (n ~ 10-30 atoms3) may only be
possible using the basic theory (i.e., Hartree Fock approximation) with minimal basis
sets, whereas calculations on chemical reactions between simple diatomic molecules
can be performed with state-of the-art model chemistries. For very large systems (n ~
50 atoms or more), timely results are only possible with less sophisticated models, like
those described in sections 2.2 and 2.3 below.

2.2 Semi-empirical methods

Semi-empirical methods increase the speed of computation by using approximations of

ab initio techniques (e.g., by limiting choices of molecular orbitals or considering only
valence electrons) which have been fitted to experimental data (for instance, structures
and formation energies of organic molecules). Until recently, the size of many energetic
molecules placed them beyond the scope of ab initio calculations, so preliminary
theoretical studies were performed using semi-empirical techniques [3-11]. However,
semi-empirical methods have been calibrated to typical organic or biological systems
and tend to be inaccurate for problems involving hydrogen-bonding, chemical
transitions or nitrated compounds [5,7,12,13].

Several semi-empirical methods are available and appear in commercially available

computational chemistry software packages such as HyperChem [14] and Chem3D
[15]. Some of the more common semi-empirical methods can be grouped according to
their treatment of electron-electron interactions [14].

2.2.1 The extended Hückel method

Extended Hückel calculations neglect all electron-electron interactions, making them

computationally fast but not very accurate. The model provides a qualitative estimate
of the shapes and relative energies of molecular orbitals, and approximates the spatial
distribution of electron density. Extended Hückel models are good for chemical
visualisation and can be applied to ‘frontier orbital’ treatments of chemical reactivity.

2.2.2 Neglect of differential overlap (NDO)

NDO models neglect some but not all of the electron-electron interactions. The Hartree-
Fock Self-Consistent Field (HF-SCF) method (see section 3.2) is used to solve the
Schrödinger equation with various approximations:

• Complete NDO (CNDO) – the product of two atomic orbitals on different atoms is
set equal to zero everywhere. In this case, repulsion between electrons in
different orbitals depends only upon the nature of the atoms involved, and not

3 Obviously, n is constantly increasing as computing power is improved.

4
 DSTO-GD-0253

on the particular orbitals. Because CNDO neglects almost all description of

electron exchange properties, it does not differentiate between states that have
the same electronic configuration, but different values of electron spin.
• Intermediate NDO (INDO) – differential overlap between orbitals on the same
atom are taken into account in the description of electron-electron repulsion, but
differential overlap between orbitals on different atoms is neglected.
• Modified INDO, version 3 (MINDO/3) – reparameterised version of INDO
optimised to predict good enthalpies of formation and reasonable molecular
geometries for a range of chemical systems, in particular, sulphur-containing
compounds, carbocations, and polynitro organic compounds [16].
• Zerner’s INDO methods (ZINDO/1 and ZINDO/S) – Michael Zerner’s (University
of Florida) versions of INDO developed for use with molecular systems
containing transition metals. ZINDO/1 is optimised to predict molecular
geometries and ZINDO/S is optimised to predict UV spectra.

2.2.3 Neglect of diatomic differential overlap (NDDO)

NDDO methods build upon the INDO model by including the overlap density
between two orbitals on one atom interacting with the overlap density between two
orbitals on the same or another atom.

• Modified NDO (MNDO) – a method introduced to correct some of the problems

associated with MINDO/3 [14]. MNDO does not work well for sterically
crowded molecules, four-membered rings, hydrogen bonding, hypervalent
compounds, nitro groups and peroxides. In general, MNDO overestimates
activation barriers to chemical reactions [15].
• Austin Method, version 1 (AM1) – a reparameterised version of MNDO which
includes changes in nuclear repulsion terms [15]. Although more accurate than
MNDO, AM1 does not handle phosphorus-oxygen bonds, nitro compounds and
peroxide bonds [14].
• Parameterisation Model, version 3 (PM3) – a second reparameterisation of MNDO,
functionally similar to AM1, but with some significant improvements. PM3 is a
recently developed semi-empirical method that may contain as yet undiscovered
defects [15].

2.3 Molecular mechanics

Molecular mechanics (MM) is often the only feasible means with which to model very
large and non-symmetric chemical systems such as proteins or polymers. Molecular
mechanics is a purely empirical method that neglects explicit treatment of electrons,
relying instead upon the laws of classical physics to predict the chemical properties of

5
DSTO-GD-0253

molecules. As a result, MM calculations cannot deal with problems such as bond

breaking or formation, where electronic or quantum effects dominate. Furthermore,
MM models are wholly system-dependent; MM energy predictions tend to be
meaningless as absolute quantities, and are generally useful only for comparative
studies. Despite these shortcomings, MM bridges the gap between quantum and
continuum mechanics, and has been used quite extensively to study ‘mesoscopic’
effects in energetic materials. Applications include modelling reaction and dissociation
on classical potential energy surfaces [17-20], studies of equilibrium crystal properties
(e.g., density, packing, specific heats) [21-27], dynamic investigations of shock
interactions with crystals and defects [28,29], and simulating detonation in molecular
crystals [30-33].

The basic assumptions of typical molecular mechanics methods are listed below.

• Each atom (i.e., electrons and nucleus) is represented as one particle with a
characteristic mass.
• A chemical bond is represented as a ‘spring,’ with a characteristic force constant
determined by the potential energy of interaction between the two participating
atoms. Potential energy functions can describe intramolecular bond stretching,
bending and torsion, or intermolecular phenomena such as electrostatic
interactions or van der Waals forces.
• The potential energy functions rely on empirically derived parameters obtained
from experiments or from other calculations.

Current molecular mechanics models are characterised by the set of potential energy
functions used to describe the chemical forces. These force fields depend upon:

• Atomic displacements (i.e., bond lengths);

• Atom types, that is, the characteristics of an element within a specific chemical
context (e.g., a carbonyl carbon versus a methyl carbon); and
• One or more parameter sets relating atom types and bond characteristics to
empirical data.

Examples of MM force fields in common use are:

• AMBER (Assisted Model Building with Energy Refinement) – primarily designed for
the study of biomolecules such as proteins and nucleotides [34].
• CHARMM (Chemistry at HARvard Molecular Mechanics) – primarily designed for
biological and pharmaceutical study, but has also been applied to micelles and
self-assembling macromolecules [35].

6
 DSTO-GD-0253

• MMx (MM2, MM3, etc) – optimised for structural and thermodynamic studies of
small non-polar molecules [14]. MMx force fields include third- and fourth-order
corrections to standard quadratic fits for the potential energy surfaces of bonds
and bond angles, thus allowing for non-harmonic effects in molecular
vibrations. The various MMx versions differ primarily in their
parameterisations. The higher versions tend to be more modern and address
deficiencies in their predecessors. However, for the newer versions such as
MM4, parameters may not be available for all classes of molecules.
• OPLS (Optimised Potentials for Liquid Simulations) – optimised for reproducing the
physical properties of biomolecules in liquid solutions [14].

The authors are not aware of any published study discussing the use of AMBER,
CHARMM, MMx or OPLS with energetic materials. Since these packages are optimised
for biochemistry and pharmaceutical applications, it is unlikely that they will
accurately reproduce the behaviour of energetic materials without further
modification. However, it is likely they can be use for limited applications with only
slight modification.
Basic techniques of molecular dynamics are described in texts such as Rapaport [36]
and Haile [37], and a general review of molecular mechanics applications in several
areas of chemistry is given by Rappé [38].

2.4 Hybrid methods

It is worth mentioning here recently developed hybrid methods which combine high-
level quantum mechanical calculations on a small part of a system with a lower-level
method on the rest of the system. Thus for large clusters or macromolecules, accurate
calculations can be carried out on the area of interest without ignoring or making
unnecessary assumptions about the remainder of the system.

Three methods have been developed and extensively used by Morokuma et al. at
Emory University (Atlanta, GA). These methods and their applications are likely to
receive increasing attention in the next few years:
• Integrated Molecular Orbital + Molecular Mechanics (IMOMM) - a two-layer method
in which a high level MO calculation is combined with molecular mechanics [39].
For example, the IMOMM method has been used to investigate organometallic-
catalysed polymerisation reactions [40,41].
• Integrated Molecular Orbital + Molecular Orbital (IMOMO) – a two-layer method
which combines high level and low-level MO calculations [42,43]. This method can
be used for calculating bond dissociation energies of large molecules [44].
• “Our own N-layered Integrated molecular Orbital molecular Mechanics (ONIOM)
method” - a technique that has largely superseded IMOMM and IMOMO, which are

7
DSTO-GD-0253

in fact, subsets of ONIOM. ONIOM is a three-level general method, allowing the

combination of any high-level method with any low- or ‘medium’-level method
[45]. ONIOM has been applied to calculations of bond breaking in fullerenes [46].

3. Ab initio Molecular Orbital Theory

What follows is a brief description of the fundamental concepts that underlie the
algorithms of typical off-the-shelf ab initio molecular orbital software. More
comprehensive discussion of molecular orbital theory may be found in selected
references [12,13,16,47-49].

3.1 The physico-chemical model

The basis of electronic structure methods is the assumption that all chemistry can be
described in terms of the interactions between electronic charges within molecules.
Hence, chemical bonds can be loosely defined as a redistribution of electronic charge that
stabilises the molecule with respect to a collection of its (isolated) constituent atoms.

Relative stabilities are expressed in terms of the total energy of the system, which is
defined by a differential equation,
Ĥ = T̂ + V̂ , (3.1.1)

where H is the Hamiltonian operator representing the sum of kinetic ( T̂ ) and potential
( V̂ ) energies. In quantum mechanical systems, the kinetic energy of a particle is
− h2 2
T̂ = ∇ , (3.1.2)
2m
where m is the mass of the particle, h is Planck’s constant ( h = 1.055×10-34 J⋅s), and
∂2 ∂2 ∂2
∇2 = + + . (3.1.3)
∂x 2 ∂y 2 ∂z 2

q1 r2 – r1
q2

r1
r2
origin

Figure 3.1 Vector diagram of pairwise electronic charge interactions.

8
 DSTO-GD-0253

For electrostatic systems, the potential energy is generally expressed in terms of

pairwise interactions between charged particles (see Fig. 3.1):

q1q 2 1
V̂ = , (3.1.4)
4πε 0 r2 − r1

where ε0 is the permittivity of free space (ε0 = 8.854×10-12 C2/N⋅m2), and |r2 - r1| is the
distance between charges q1 and q2.

3.1.1 The molecular Hamiltonian

Under the basic assumption of electronic structure methods, a molecule is a collection

of charged quantum particles. The molecular Hamiltonian has the form of Eq. 3.1.1
above, however the kinetic energy is now a summation over all the particles in the
molecule,

− h2 1  ∂2 ∂2 ∂2 
T̂ = ∑  2 + 2 + 2  , (3.1.5)
2m i mi  ∂xi ∂y i ∂z i 

and the potential energy is the Coulomb interaction between each pair of charged
particles (electron-nucleus attraction, nucleus-nucleus repulsion, and electron-electron
repulsion):

1 q j qk
V̂ = ∑∑ , (3.1.6)
4πε 0 j k < j rk − r j

For electrons, q = -e (e = 1.602×10-19 C), and for nuclei of atomic number Z, q = +Ze.

3.2 Hartree-Fock theory

3.2.1 The Schrödinger equation

The quantum mechanical description of chemical bonds is given by a space- and time-
dependent probability distribution: the molecular wavefunction, Ψmol(t). The molecular
wavefunction is defined by the Schrödinger equation

∂Ψmol (t )
Ĥ mol Ψmol (t ) = ih . (3.2.1)
∂t

If the potential energy operator is time-independent, then the solution obtained by

separation of variables, leads to the molecular wavefunction

9
DSTO-GD-0253

Ψmol (t ) = ψ mol e − iEmol t / h , (3.2.2)

where ψmol satisfies the time-independent Schrödinger equation

Ĥ mol ψ mol = E mol ψ mol , (3.2.3)

and Emol is the total energy of the molecule. Solutions of the time-independent
Schrödinger equation represent various stationary states of the molecule
(corresponding to stable or meta-stable electronic configurations). The set of
wavefunctions ψ which satisfy Eq. 3.2.3 are its eigenfunctions, and the energies of the
molecule, Emol, in each stationary state are its eigenvalues. The stationary state with the
lowest energy is called the ‘ground state.’

3.2.2 Antisymmetry and electron spin

Standard electronic structure methods assume that the molecular wavefunction

describing several electrons can be written as a product of single-electron
wavefunctions called ‘orbitals’; that is, for a molecule containing n electrons,

ψ mol (1,2,...n) = ψ (1)ψ (2)...ψ (n) . (3.2.4)

Electrons possess an intrinsic angular momentum or ‘spin’ with a value of ±½. A half-
integer spin quantum number implies that electrons are antisymmetric with respect to
exchange – in other words, a wavefunction describing a pair of electrons i and j must
change sign when the electrons are interchanged:

ψ(i, j ) = −ψ ( j , i ). (3.2.5)

The simplest antisymmetric combination of molecular orbitals (MOs) is a matrix

determinant. A HF wavefunction is constructed by assigning electrons to molecular
orbitals φ(r) in pairs of opposite spin, and then forming a determinant using two
spin functions α and β, where

α(↑) = 1 α(↓) = 0 (3.2.6)

β(↑) = 0 β(↓) = 1. (3.2.7)

For two electrons i and j, the total wavefunction takes the form:

α(i ) β(i )
ψ(i, j ) = φ(r ) (3.2.8)
α ( j ) β( j )

with a determinant

10
 DSTO-GD-0253

φ(r )
ψ(i, j ) = [α(i)β( j ) − β(i)α( j )] (3.2.9)
2
which satisfies the antisymmetrisation condition of Eq. 3.2.4. For a molecule containing
n electrons, the wavefunction is referred to as a ‘Slater determinant,’ and takes the
form:

φ1 (1)α(1) φ1 (1)β(1) φ 2 (1)α(1) φ 2 (1)β(1) L φ n (1)α(1) φ n (1)β(1)

2 2

1 φ1 (2)α(2) φ1 (2)β(2) φ 2 (2)α(2) φ 2 (2)β(2) L φ n 2 (2)α(2) φ n 2 (2)β(2)

ψ mol =
n! M M M M M M
φ1 (n)α(n) φ1 (n)β(n) φ 2 (n)α(n) φ 2 (n)β(n) L φ n (n)α(n) φ n (n)β(n)
2 2

(3.2.10)

3.2.3 Ab initio essentials

For systems of more than two interacting particles, the Schrödinger equation cannot be
solved exactly.4 Therefore, all ab initio calculations for many-body systems
(e.g., molecules) involve some level of approximation and indeed, some level of
empirical parameterisation.5 Nevertheless, ab initio methods for molecular calculations
must satisfy a set of stringent criteria:

1. Solutions must be well defined and specified by both the structure and the
electronic states of the molecule.
2. The potential energy of the molecule must vary smoothly and continuously with
respect to displacements of the atomic nuclei.
3. The model must contain no bias (e.g., assuming a chemical bond exists between two
atoms.)
4. The model must be ‘size-consistent’ – that is, solutions and their associated errors
must scale in proportion to the size of the molecule.6
5. The model must be ‘variational’ – that is, approximate solutions must provide an
upper bound to the ‘true’ energy of the system. Consequently, the approximate
solution having the lowest energy represents the closest fit to the true
wavefunction, within the constraints of the method.

4 This is true for any system, quantum or classical, since current permutation operators (e.g., the
Levi-Civita symbol, εijk) can only account for pairwise (two-particle) relationships [13].
5 Indeed, ab initio electronic structure calculations could be considered the ‘ultimate’ semi-

empirical approximation! It is perhaps more consistent to describe ab initio methods as

‘universal,’ that is, the a set of wavefunctions are optimised for a particular method using
empirical data from a set of atoms and molecules, and it is then assumed that these
wavefunctions may be applied to calculations involving all atoms and molecules.
6 Size-consistency allows direct comparison between molecules of different sizes, and is

therefore an important consideration for studies of chemical reactions.

11
DSTO-GD-0253

3.2.4 Approximations

3.2.4.1 Non-relativistic solutions

Typical ab initio electronic structure methods are time-independent, solving only the
spatial wavefunction, ψ(r). This approximation is reasonable for most chemical
systems, but breaks down in calculations involving large atoms (late transition metals,
lanthanides, etc.), whose inner-shell electrons have velocities approaching the speed of
light.

3.2.4.2 Born-Oppenheimer approximation

Electrons in molecules are much lighter than nuclei, and therefore generally have much
higher velocities. Hence, under most circumstances, one can assume that electrons
respond instantaneously to nuclear displacements. In practice, this means that the
molecular Hamiltonian can be written assuming the nuclear positions are fixed (i.e.,
neglecting nuclear kinetic energy terms):

− h2 electrons 1 q j qk
Ĥ mol = T̂elec + V = ∑ ∇ i2 + ∑∑ . (3.2.11)
2 me i 4πε 0 j k > j rk − r j

Most ab initio calculations solve only the electronic part of the molecular wavefunction,
and therefore cannot account for systems where the electronic states are strongly
coupled to nuclear vibrations.

3.2.4.3 Single particle approximation

As mentioned in Section 3.2.2, standard electronic structure methods approximate the

total wavefunction of a many-electron system as the product of single-electron
wavefunctions (Eq. 3.2.4). This is the essence of Hartree-Fock (HF) theory, which
describes each electron in a molecule as moving in the average electric field generated
by the other electrons and nuclei. As a single-particle theory, HF theory systematically
overestimates molecular energies because it neglects the correlated motion of electrons
resulting from Coulomb interactions.

3.2.4.4 Linear combinations of atomic orbitals (LCAO)

Although there is no exact analytical solution to the time-independent molecular

Schrödinger equation (Eq. 3.2.3) for systems containing more than one electron,
approximate solutions can be obtained using standard numerical techniques. The
approach of all ab initio techniques is to build the total wavefunction from a ‘basis’ set
of mathematical functions capable of reproducing critical properties of the system. An
individual molecular orbital may then be expressed as

12
 DSTO-GD-0253

N
φ i (r ) = ∑ cµi χ µ (r ) , (3.2.12)
µ =i

where χµ(r) are the basis functions, and the coefficients cµi are adjustable parameters.
For a molecular wavefunction, the electronic orbitals of the constituent atoms form a
natural set of basis functions. These atomic orbitals can in turn be represented by
different types of mathematical functions. A highly accurate set of atomic orbitals
(Slater-type orbitals or STOs) are based on hydrogenic wavefunctions having the form

χ STO (r ) ~ Ce − αr . (3.2.13)

Exponential functions are not well suited to numerical manipulation, so most electronic
structure calculations approximate STOs with a linear combination of gaussian-type
functions,

χ STO (r ) ~ χ µ = ∑ d µν e − α ν r ,
2
(3.2.14)
ν

where dµν and αν are adjustable parameters. As can be seen from Fig. 3.2, gaussian-type
functions provide reasonable approximations of STOs, except at very small or very
large electron-nucleus separations.

(a) 1 (b) 1

f STO −3G ( x) = 14 e −5 x
2

0.75 0.75
+ 13 e − 20 x + 13 e −150 x
2 2

f G ( x ) = e −5 x
2

0.5 0.5

0.25 0.25
f STO ( x) = e −5 x

0 0
0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1

Figure 3.2 (a) Comparison of exponential and gaussian functions. (b) Comparison of the same
exponential function and a sum of three gaussians.

Linear combinations of ‘primitive’ gaussian functions are referred to as ‘contracted’

gaussians. Standard ab initio software packages like Gaussian98 [50] offer a choice of
basis sets containing contracted gaussians optimised to reproduce the chemistry of a
large range of molecular systems. A more detailed discussion of molecular basis sets is
provided in section 3.3.

13
DSTO-GD-0253

3.2.5 The variational principle and Roothan-Hall equations.

The sections above describe a method for constructing a determinantal wavefunction

from a set of LCAO-MOs. It remains now to define a method to determine the MO
expansion coefficients cµi (Eq. 3.2.11) which optimise the molecular wavefunction for a
particular system. The variational nature of the model (Section 3.2.3) guarantees that
the energy eigenvalue solution for any approximate wavefunction is always greater
than the energy obtained using the exact wavefunction. It follows that the set of
coefficients which minimise the energy of the resultant wavefunction will give the best
approximation to the exact wavefunction from a chosen basis set.

The variational constraint leads to a set of algebraic equations (Roothan-Hall) for cµi,
expressed in matrix form as

FC = SCε (3.2.15)

where
• C is the matrix of MO expansion coefficients;
• F is the Fock matrix, which is the sum of a term representing the energy of a
single electron in the field of the bare atomic nuclei and a term describing
electron-electron repulsion within an averaged field of electron density;
• S is a matrix describing the overlap of molecular orbitals; and
• ε is a diagonal matrix containing the one-electron energies of each molecular
orbital χµ.

Since the terms within the Fock matrix F depend upon the electron density, which in
turn, depends upon molecular wavefunction defined by the matrix of MO expansion
coefficients C, the Roothan-Hall equations are nonlinear, and must be solved by an
iterative procedure termed the ‘self-consistent field’ (SCF) method. Upon convergence
of the SCF method, the minimum-energy MOs produce the electric field which
generate the same orbitals (hence, the self-consistency).

3.3 Basis Sets

In general, a basis set is an assortment of mathematical functions used to solve a

differential equation. In quantum chemical calculations, the term ‘basis set’ is applied
to a collection of contracted gaussians representing atomic orbitals, which are
optimised to reproduce the desired chemical properties of a system.

Standard ab initio software packages generally provide a choice of basis sets that vary
both in size and in their description of the electrons in different atomic orbitals. Larger
basis sets include more and a greater range of basis functions. Therefore, larger basis

14
 DSTO-GD-0253

sets can better refine the approximation to the ‘true’ molecular wavefunction7, but
require correspondingly more computer resources. Alternatively, accurate
wavefunctions may be obtained from different treatments of electrons in atoms. For
instance, molecules containing large atoms (Z > 30) are often modelled using basis sets
incorporating approximate treatments of inner-shell electrons which account for
relativistic phenomena.

‘Minimal’ basis sets contain the minimum number of AO basis functions needed to
describe each atom (e.g., 1s for H and He; 1s, 2s, 2px, 2py, 2pz for Li to Ne). An example
of a minimal basis set is STO-3G, which uses three gaussian-type functions (3G) per
basis function to approximate the atomic Slater-type orbitals (see Fig. 3.2b). Although
minimal basis sets are not recommended for consistent and accurate predictions of
molecular energies, their simple structure provides a good tool for visualising
qualitative aspects of chemical bonding. Improvements on minimal basis sets are
described below and illustrated in Fig. 3.3.

3.3.1 Split valence basis sets

In split valence basis sets, additional basis functions (one contracted gaussian plus
some primitive gaussians) are allocated to each valence atomic orbital. The resultant
linear combination allows the atomic orbitals to adjust independently for a given
molecular environment. Split valence basis sets are characterised by the number of
functions assigned to valence orbitals. ‘Double zeta’ basis sets use two basis functions
to describe valence electrons, ‘triple zeta’ use three functions, and so forth. Basis sets
developed by Pople and coworkers [51] are denoted by the number of gaussian
functions used to describe inner and outer shell electrons. Thus ‘6-21G’ describes an
inner shell atomic orbital with a contracted gaussian composed of six primitive
gaussians, an inner valence shell with a contracted gaussian composed of two
primitives, and an outer valence shell with one primitive. Other split-valence sets
include 3-21G, 4-31G, and 6-311G.

Split-valence orbitals Polarisation functions

+ = + =
p p'′ p d

Figure 3.3. Basis set improvements.

7 It must be emphasised that although MOs are used to obtain values for physical ‘observables’
(e.g., molecular heats of formation, electronic dipole moments), they are strictly mathematical
constructs (indeed, they are solutions to the Hartree-Fock approximation as opposed to the true
many-body electronic Hamiltonian), and do not themselves contain any physical meaning.

15
DSTO-GD-0253

3.3.2 Polarised basis sets

Polarisation functions can be added to basis sets to allow for non-uniform

displacement of charge away from atomic nuclei, thereby improving descriptions of
chemical bonding. Polarisation functions describe orbitals of higher angular
momentum quantum number than those required for the isolated atom (e.g., p-type
functions for H and He, and d-type functions for atoms with Z > 2), and are added to
the valence electron shells. For example, the 6-31G(d) basis set is constructed by adding
six d-type gaussian primitives to the 6-31G description of each non-hydrogen atom.
The 6-31G(d,p) is identical to 6-31G(d) for heavy atoms, but adds a set of gaussian p-
type functions to hydrogen and helium atoms. The addition of p-orbitals to hydrogen is
particularly important in systems where hydrogen is a bridging atom.

3.3.3 Diffuse basis sets

Species with significant electron density far removed from the nuclear centres
(e.g., anions, lone pairs and excited states) require diffuse functions to account for the
outermost weakly bound electrons. Diffuse basis sets are recommended for
calculations of electron affinities, proton affinities, inversion barriers and bond angles
in anions. The addition of diffuse s- and p-type gaussian functions to non-hydrogen
atoms is denoted by a plus sign – as in ‘3-21+G.’ Further addition of diffuse functions
to both hydrogen and larger atoms is indicated by a double plus8.

3.3.4 High angular momentum basis sets

Basis sets with multiple polarisation functions are now practical for many systems and,
although not generally required for Hartree-Fock calculations, are useful for describing
the interactions between electrons in electron correlation methods (see section 3.4).
Examples of high angular momentum basis sets include:
• 6-31G(2d) - two d-functions are added to heavy atoms;
• 6-311G(2df, pd) – besides the (311) valence functions, two d functions and one f
function are added to heavy atoms, and p and d functions to hydrogen;
• 6-311G(3df, 2df, p) - three d functions and one f function are added to atoms with
Z > 11, two d functions and one f function to first-row atoms (Li to Ne) and one p
function to hydrogens.
High angular momentum basis sets augmented with diffuse functions represent the
most sophisticated basis sets available. As is discussed in Section 4.3.3 below, energetic
molecules have a complicated electronic structure due to the presence of nitro groups.
Therefore, the most accurate ab initio studies of energetic materials would be produced
by reasonably sophisticated polarised split-valence basis sets augmented with high
angular momentum and diffuse atomic orbitals. However, the size of the optimum
basis set, especially when used with electron correlation methods (discussed in

8 Diffuse functions on hydrogen atoms seldom make a difference in accuracy [1].

16
 DSTO-GD-0253

Section 3.3 below) will ultimately be determined by the size of the energetic molecule,
the amount computing power available, and the time allotted for the studies. For
example, conformational studies of RDX (1,3,5 trinitro-1,3,5-triazacyclohexane) range
from rudimentary HF/STO-3G calculations performed in 1991 [52] to more
sophisticated investigations in 1997 using B3LYP density functional theory and
6-311G(d,p) [53].

3.4 Electron correlation

Hartree-Fock theory is a single-particle approximation, and therefore cannot

adequately treat the correlated motion of electrons that occurs due to electron-electron
interactions. Neglect of electron correlation has been blamed for systematic HF errors
such as underestimated bond lengths and overestimated vibrational frequencies.
Calculations added to HF-SCF theory to remedy these errors are termed ‘electron
correlation’ or ‘post-HF’ methods. There are three general types of electron correlation
treatments: configuration-interaction (CI) methods, Møller-Plesset (MP) perturbation
theory, and density functional theory (DFT).

3.4.1 Configuration-interaction

A HF wavefunction of a molecule has one determinant; therefore, it can only describe

only one electronic configuration in a molecule. This approximation is so restrictive
that HF theory cannot predict dissociation of the simplest molecule, the hydrogen
dimer, into its neutral atomic components. However, correct dissociation behaviour is
restored when extra configurations, corresponding to electronically excited states, are
added to the wavefunction. Configuration-interaction methods incorporate excited
state configurations into the wavefunction by constructing new determinants from the
original HF determinant. New determinants are created by replacing one or more
occupied orbitals with unoccupied (virtual) orbitals of higher energy. The number of
replacements within the determinants designates the level of CI. For instance, single
substitution (CIS) switches one pair of occupied and virtual orbitals,

ψ HF = φ1, ...φ j ,...φ HOMO , φ LUMO ,...φ i ,...φ n →

ψ CIS = φ1, ...φ i ,...φ HOMO , φ LUMO ,...φ j ,...φ n (3.4.1)

and is equivalent to a one-electron excitation. Higher-order calculations include CID

(double substitutions), which generates determinants where two orbital pairs are
switched; CISD, which adds singly and double-substituted determinants; and CISD(T),
with single, double and triple excitations. The theoretical limit of this expansion - a full
CI calculation - forms the molecular wavefunction as a linear combination of the HF
determinant and all possible substituted determinants:

17
DSTO-GD-0253

ψ full −CI = a 0 ψ HF + ∑ a n ψ S , D ,T ,... (3.4.2)

n>0

where the coefficients ai are determined by minimising the energy of the total
wavefunction. Full CI provides the most complete non-relativistic treatment possible
for a molecular system; however, it is extremely computer-intensive, and generally
untenable for all but the simplest molecules described by small basis sets.

3.4.2 Møller-Plesset perturbation

For typical molecules, most of the ground-state energy originates from one-electron HF
contributions. Møller-Plesset perturbation theory assumes that the effects of electron
correlation are minor, and can be described by small corrections (perturbations) to the
HF solution. Essentially, MP methods assume the true molecular Hamiltonian (3.2.3)
can be divided into two parts:

Ĥ mol = Ĥ one − e + λ P̂many −e (3.4.3)

where Ĥ one− e denotes single-electron energy contributions that can be solved exactly by
HF-SCF, and P̂many −e represents contributions due to electron correlation. The
coefficient λ is used to generate power series expansions of the energy and the
molecular wavefunction:

E mol = E ( 0) + λE (1) + λ2 E ( 2 ) + λ3 E (3) + ... (3.4.4)

ψ mol = ψ (0)
+ λψ (1)
+λ ψ 2 (2)
+λ ψ
3 ( 3)
+ ... (3.4.5)

These series are then substituted back into the molecular Schrödinger equation with
the modified Hamiltonian (3.4.3), and each term is evaluated in turn. The first two
energy terms constitute the HF energy

E HF = E ( 0) + E (1) , (3.4.6)

and are used to evaluate ψ(1), which is composed of a linear combination of substituted
determinants

ψ (1) = ∑ bn ψ S , D ,T ,... , (3.4.7)

n

with coefficients bi that are inversely proportional to the difference in energy between
the ground state and the associated excited states. That is, substituted wavefunctions
close in energy to the ground state make larger contributions to the perturbation
expansion. The first-order correction to the wavefunction, ψ(1), is then used to calculate
the second-order correction to the total energy E(2), and so on, back and forth, until the
desired order of correction is achieved. Møller-Plesset theory is designated by the order

18
 DSTO-GD-0253

of the perturbative corrections. Since the second-order energy term E(2) is the first-order
correction to the HF energy, the designation begins at MP2, and continues with MP3,
MP4, MP5, and so forth.

3.4.3 Density-functional theory

DFT theory models electron correlation as a functional9 of the electron density, ρ. The
functional employed by current DFT methods partitions the electronic energy via the
Kohn-Sham equations [54,55] into several terms

E = E T + E V + E J + E XC (3.4.8)

where ET is the kinetic energy term (arising from the motion of the electrons), EV is the
potential energy term that includes nuclear-electron and nuclear-nuclear interactions, EJ
is the electron-electron repulsion term and EXC is the electron correlation term. All terms
except nuclear-nuclear repulsions are functions of the electron density. The terms ET +
EV + EJ represent the classical energy of the electron distribution, while EXC represents
both the quantum mechanical exchange energy, which accounts for electron spin10, and
the dynamic correlation energy due to the concerted motion of individual electrons.
Pure DFT methods calculate EXC by pairing an exchange functional with a correlation
functional and so are designated by the choice of combination. For example, BLYP
combines Becke’s gradient-corrected exchange functional with the gradient-corrected
correlation functional of Lee, Yang and Parr [56].

DFT calculations fall into three general categories: local density approximations (LDA),
generalised gradient approximations (GGA), and ‘hybrid’ combinations of DFT and
Hartree-Fock terms. LDA exchange and correlation functionals only contain terms
related to electron density11 - an approach that works for some bulk materials, but fails
to accurately predict properties in isolated molecules. GGA (‘nonlocal’) functionals
contain terms that depend upon both the electron density and the density gradients12.

9 A functional is a function defined by another function, that is, a function of a function.

10 The quantum-mechanical exchange energy is accounted for in HF theory by generating
antisymmetric determinental wavefuntions, as discussed in section 2.2.2. Thus, Coulomb “self-
interaction” energy can be considered a type of electron correlation already included in HF
theory.
11 An example of a local functional is the ‘Slater’ functional:
r
= − 32 (43π ) ∫ρ
XC 1/ 3 4/3
E LDA dr
v
r
where ρ, the electron density, is a function of r . This local functional reproduces the properties
of a uniform electron gas. Note: this is not the same as Slater determinants discussed in the
previous footnote.
12Gradients represent spatial changes of a function as can be quite large. For example, the

electronic density within a metal might be quite uniform, and thus be well modelled by the
Slater functional in footnote 6, yet display large gradients at the metal surface. Large gradients

19
DSTO-GD-0253

The gradient-corrected density functional method BLYP is capable of predicting

intramolecular bond dissociation energies to within a few kJ/mol [57]. However, the
generalised gradient approximation severely underestimates activation barriers for
some reactions due to neglect of Coulomb ‘self-interaction’ of the electrons [58]. This
problem is remedied with hybrid methods that combine Hartree-Fock self-interaction
corrections with density functional exchange and correlation. Examples of hybrid
methods are B3LYP and B3PW91, where B3 denotes Becke’s three-parameter hybrid
functional [59], while ‘PW91’ and ‘LYP’ are gradient-corrected correlation functionals of
Perdew and Wang [60] and, as above, Lee, Yang and Parr.

3.4.4 Caveats for post-HF methods

Although post-SCF treatments should improve upon Hartree-Fock theory, each of

these techniques violates at least one of the fundamental criteria that define ab initio
methodology. Truncated CI calculations (i.e., anything less than full configuration
interaction) are not size-consistent, so comparative CI studies should include added
corrections13. MP perturbation theory is not a variational technique, and as such does
not provide an upper bound to the true molecular energy14. Finally, DFT contains a
bias: the exact functional dependence of EXC upon the electron density is unknown and
must be approximated by assuming that ρ has certain properties (e.g., it behaves like a
uniform electron gas). Excited-state DFT calculations, which require the application of
linear response theory, has so far been limited to atomic systems15. These caveats serve
as a reminder that although ab initio methods are capable of yielding predictions close
to experimental results, many of the approximations that make quantum chemical
computations feasible are wholly nonphysical. Hence, even the most sophisticated
ab initio software packages should not be treated simply as ‘black boxes.’

Given these considerations, the results of any MO calculation on energetic materials are
subject to scrutiny. Nevertheless, one can assume with a high level of confidence that
DFT/6-31G(d) will yield bond lengths accurate to ±0.01Å, and the molecular
vibrational frequencies to within ±100 cm-1. However, DFT predictions of chemical
reaction energies should be treated as qualitatively correct at best. For quantitative

also exist across interfaces of p- and n- type semiconductors, and govern gating and switching
of electric current in transistors.
13 One method , quadratic configuration interaction (QCI), restores size consistency by adding

terms, with corrections available up to quadruply excited states: QCISD, QCISD(T) and
QCISD(TQ). Coupled cluster methods correct for size-consistency by accounting for
simultaneous substitutions in different molecules – a technique useful for modelling chemical
reactions. Coupled-cluster calculations (CCD, CCSD) are useful when electron correlation
makes a major contribution to electron-pair bonds and so cannot be reasonably described by
perturbation methods [61].
14 The second-order term (MP2) is always negative and often overcorrects for electron

correlation.
15 In other words, DFT should not be used to calculate excited states of molecules with existing

commercial software [53].

20
 DSTO-GD-0253

accuracy (± 40 kJ/mol), a recommended procedure is to optimise the molecular

structure using DFT with a high-level split-valence polarised basis set, then recalculate
the structural energy with MP4 and the same or a slightly less sophisticated split-
valence polarised basis set.

3.5 Open-shell calculations and spin contamination

In addition to the considerations above, input to molecular orbital calculations must

specify the treatment of electron spin - that is, whether the molecule will be described

ψ4 ψ4β
ψ4α
ψ3 ψ3β
E
ψ3α
ψ2 ψ2β
ψ2α
ψ1 ψ1β
ψ1α

Figure 3.4. Electron assignments for closed-shell and open-shell calculations.

using an ‘open-shell’ or ‘closed-shell’ model. Most molecules have singlet electronic
ground states (zero total electron spin): they contain an even number of electrons
which can be assigned to orbitals in pairs of opposite spin. HF models that pair
electrons in this way are referred to as ‘spin-restricted closed-shell’ calculations, and
are adequate for treating most types of chemical systems. However, there are several
cases which require explicit consideration of unpaired electrons, and are better
described by open-shell unrestricted models that generate separate MOs for spin-up
(α) and spin-down (β) electrons. Spin-unrestricted calculations are used for:

• Molecules with odd numbers of electrons (e.g., some ions and radicals);
• Excited states;
• Molecules with triplet or higher-spin ground states (e.g., O2);
• Processes that separate electron pairs (e.g., dissociation); and
• Delocalised orbitals for resonant systems.

Spin-unrestricted wavefunctions are not eigenfunctions of the spin operator ( Ŝ 2 ), and

therefore suffer from intermixing between spin states. In most open-shell systems, spin
contamination is negligible; however, for molecules where the unpaired electrons are
delocalised, contamination can be quite significant, especially from the next-higher
spin state. Unrestricted calculations of nitric oxide, a diatomic molecule with one
unpaired electron and a doublet (2π) ground state, display extreme spin contamination

21
DSTO-GD-0253

when small basis sets (STO-3G or 3-21G) are used16. Spin contamination can also be
expected in calculations of energetic materials containing nitro (-NO2) groups
involving excited states or radical producing reactions.

The detrimental effects of spin contamination can be appreciable, although the degree
of spin contamination decreases with increasing basis set size. Spin contamination has
been blamed for poor predictions of molecular geometries [62] and vibrational
frequencies [63]. The effects of spin contamination are particularly acute when
occupied and unoccupied orbitals are nearly degenerate. This can happen, for instance,
when a bond is stretched to the limit of dissociation, and the separation between
bonding and antibonding orbitals approaches zero, generating a wavefunction that
contains an equal mixture of singlet and triplet states. Another example is ethylene
torsion, where the π and π* orbitals become equivalent by symmetry at the transition
state for rotation about the double bond [64].

It is more desirable to remove spin contamination than to ignore it, however, spin
annihilation is beyond the scope of simpler post-HF methods. Open-shell HF
wavefunctions effectively introduce some electron correlation, because the use of
different orbitals for each electron allows α and β electrons to be spatially further apart.
Hence, simpler correlation corrections (MP2 and CIS) do not generally improve spin-
contaminated HF wavefunctions [65]. Suggested spin annihilation methods include
high levels of electron correlation, spin-projected energy corrections, or the use of
various forms of restricted open-shell wavefunctions. Gaussian 98 automatically
incorporates annihilation of one spin contaminant into its UHF calculations. Density
functional methods also appear to reduce spin contamination. Baker, Scheiner and
Andzelm [66] observed that ‘pure’ unrestricted Kohn-Sham wavefunctions display
significantly less spin contamination than their HF counterparts, and concluded that
spin contamination arises primarily from the use of Slater determinants in HF theory.
Therefore, if spin contamination is a consideration in density functional calculations,
hybrid methods that include HF exchange should be avoided.

16 An obvious manifestation of spin contamination is an overestimate of spin-squared

expectation values. For a spin eigenfunction φ(α,β), Ŝ 2 φ(α,β) = s(s+1)φ(α,β), where s is the total
spin. For pure doublet states (s = ½), < Ŝ 2 > = ½ (½ + 1) = 0.75. As-mentioned unrestricted HF
calculations of nitric oxide yield values of < Ŝ 2 > ranging from 1.0 to 1.45.

22
 DSTO-GD-0253

4. Accuracy versus Efficiency for Ab initio Methods

4.1 Error reduction

4.1.1 Methods

The standard MO treatment for most closed-shell molecules is a spin-restricted

Hartree-Fock self-consistent field calculation. HF-SCF calculations generally yield
accurate molecular structures but are less successful at predicting molecular energies.
Depending upon the system and the choice of basis set, HF energies typically fall
anywhere within 200 kJ/mol of experimental values, although errors can be as large as
700 kJ/mol. The main source of error in HF calculations is neglect of electron
correlation, which results in systematic overestimates of molecular energies. To
account for electron correlation, calculations are added to HF results, including
perturbation treatments (e.g., MP2 and MP4), multiconfiguration methods (e.g., CISD,
CCSD and CASSCF17), and density functional theory (e.g., BLYP, and B3PW91).
Perturbation methods and density functional theory (DFT) are typically accurate to
within 40 kJ/mol, while the results of multiconfiguration treatments are even better,
falling within 5-8 kJ/mol of experiment. Since multiconfiguration calculations are
extremely expensive for even moderately sized molecules, perturbation treatments and
DFT tend to be the methods of choice for treating electron correlation.

4.1.2 Basis sets

In general, ab initio MO studies on complex unknown systems should begin with

calculations using small basis sets, both to obtain a timely qualitative assessment of
molecular properties, and to determine the computational cost of the higher-order
calculations. However, minimal basis sets like STO-3G should only be used when the
size of the molecule is beyond the scope of available computer resources for even small
split-valence sets like 3-21G. Quantitative accuracy improves with the size of basis sets,
since larger basis sets contain more adjustable parameters and are thus better able to
approximate true molecular wavefunctions. Polarisation functions improve the
accuracy of fundamental basis sets (e.g., changing 6-31G to 6-31G(d)), although the
ultimate gain is limited18. Diffuse functions (e.g., changing 6-31G to 6-31+G) will
improve results under certain circumstances. For general use, the smallest standard
basis set recommended by developers of Gaussian 98 is 6-31G(d) [51].

17 CASSCF is the acronym for ‘complete active-space self-consistent field’ calculations. CASSCF
is a ‘full’ configuration-interaction calculation over a limited number of electrons. That is, CI
determinants are constructed for specific molecular orbitals, usually those containing valence
electrons [51].
18 That is, for a given property, calculations performed with the 6-311+G(d,p) may be just as

accurate as those using 6-311++G(3df,3pd), in which case 6-311+G(d,p) defines the ‘basis set
limit.’

23
DSTO-GD-0253

4.2 Resource usage

Given the limits of computer memory and speed, the size of the chemical system
generally determines the maximum size of the basis set used with MO methods. For
instance, a moderately sized energetic molecule, 1,3,3-trinitroazetidine (TNAZ),
contains 49 pairs of electrons. The 6-31G(d) basis set assigns 203 basis functions to the
molecule, 49 of which are occupied, and the remaining 154, unoccupied or ‘virtual’
orbitals. To estimate memory requirements (in 8-byte words), Gaussian 98 references
[51] suggest using
M + 2NB2 (4.1.1)
where NB is the number of basis sets used in the calculation and the value of M
depends upon the method of calculation (typically around 4-6 MW).

Table 4.1 Resource usage by method and basis set

Method CPU Memory Disk
SCF N4 N2 N4
MP2 ON4 N2 N4
MP4/QCISD(T) O3V4 N2 N4
N = total number of basis functions
O = total number of occupied orbitals
V = total number of virtual orbitals

According to Eq. 4.1.1, a Hartree-Fock SCF single-point energy calculation using the
6-31G(d) basis set (denoted ‘HF/6-31G(d)’) requires approximately 4.3 MW (34 MB)
worth of memory. On a SGI Origin 200019, the recorded CPU time for this calculation is
19 minutes. From the guidelines listed in Table 4.1, a calculation using a larger basis set
will require significantly more memory and CPU time. For instance, a HF/aug-cc-
pVTZ20 SPE, which assigns 609 basis functions to TNAZ, would require at least 40MB
of memory, and take over a day to run21. Electron correlation treatments are also
expensive: a MP2/6-31G(d) SPE would take roughly 16 hours to run, while MP4 or
QCISD(T)/6-31G(d) would take a year and a half. Clearly, sophisticated post-HF
calculations, such as CI and MP2 coupled with correlation-consistent basis sets, are
only feasible for smaller molecules displaying unusual electronic properties.

19 This calculation was run with the standard default options of Gaussian98. The default mode
imposes modest convergence criteria sufficient for 0.5 kJ/mol accuracy in the SCF energy and
three decimal place accuracy in quantities derived from the electron densities (e.g. electrostatic
potentials and derived charges). All calculations were performed under the IRIX 6.4 operating
system using one 185MHz CPU optimised for floating-point arithmetic.
20 This is the acronym for Dunning’s correlation-consistent polarised valence triple-zeta basis set

[67] augmented by diffuse functions. ‘Correlation-consistent’ refers to optimisation of the basis

sets with electron-correlated atomic calculations.
21 Recorded CPU time for single RHF/aug-cc-pVTZ energy calculation on the SGI Origin 2000: 4

days, 17 hours, 41 min and 16.7 sec.

24
 DSTO-GD-0253

Density functional methods avoid the computational expense of other post-HF

methods by approximating the effects of electron correlation. DFT calculation times
formally scale as the third power of the number of electrons in the system, Ne3, whereas
computational times for MP4 or QCISD(T) scale as Ne5. In practical terms, this means
that DFT treatments of electron correlation in TNAZ should only take four times as
long as a HF calculation22. Given its demonstrated accuracy for predicting molecular
energies and activation barriers, the enormous amount of time saved by using DFT
makes it an attractive method for studying the ground state properties and reactions of
larger energetic molecules like RDX or HMX.

4.3 High-accuracy/low-cost model chemistries

Several procedures have been developed to accurately predict23 thermochemical

quantities such as atomisation energies, ionisation potentials, electron affinities and
proton affinities, at relatively low computational cost. These procedures fall into two
general categories: compound methods, and complete basis set (CBS) models

4.3.1 Compound methods

Compound methods such as Gaussian-1 [68] and Gaussian-2 [69] (‘G1’ and ‘G2’)
attempt to approximate a single high-level calculation by a combination of several low-
level calculations. For example, the procedure for a G1 calculation is as follows:

1. Optimise the molecular structure using HF/6-31G(d) and determine the vibrational
zero-point energy (ZPE)24.
2. Reoptimise the molecular structure using full (all-electron) MP2/6-31G(d). Perform
all subsequent calculations using this geometry.
3. Compute a base-level energy (Ebase) at MP4/6-311G(d,p).
4. Compute the energy correction ∆E+ obtained by including diffuse functions
(MP4/6-311+G(d,p)): ∆E+ = Ebase+diff – Ebase.
5. Compute the energy correction ∆E2df obtained by including higher polarisation
functions on atoms with Z >2 (MP4/6-311G(2df,p): ∆E2df = Ebase+2df – Ebase.
6. Compute the energy correction for residual electron correlation effects (to
counteract the known deficiencies of truncating perturbation theory to fourth
order) using QCISD(T)/6-311G(d,p): ∆EQCI = EQCISD(T) – Ebase.

22 Actual test time for single B3LYP/6-31G(d) energy calculation on the SGI Origin 2000: 34 min
18 sec.
23 In some cases, predictions lie within 1% of experimental values.
24 The ZPE is generally corrected by a scale factor which accounts for known deficiencies in the

harmonic approximation of intramolecular potential energy surfaces. For HF/6-31G(d), the

scale factor is 0.8929.

25
DSTO-GD-0253

7. Estimate a higher-level correction of the remaining correlation energy between

spin-paired electrons using the formula ∆EHLC = -0.00019nα + -0.00595nβ, where nα
is the number of spin-up electrons and nβ is the number of spin-down electrons in
the molecule.

The resulting G1 energy, EG1 = Ebase + ∆E+ + ∆E2df + ∆EQCI + ∆EHLC + ZPE, is essentially
an approximation to the energy of a much more computationally intensive calculation,
QCISD(T)/6-311+G(2df,p).

The G2 procedure includes a couple of corrections to the G1 energy, and, for molecules
containing first- and second-row atoms, is reported to predict properties to within an
average absolute deviation of 6.7 kJ/mol of experimental results. A modified version of
G2 known as G2(MP2) uses second-order instead of fourth-order perturbation theory,
providing nearly the same accuracy as full G2 at substantially lower computational
cost [1].

G1 and G2 methods have been used to study the smaller energetic molecules,
nitromethane (H3C-NO2) [70, 71] and nitramide (H2N-NO2) [72]. For nitromethane,
heats of formation and gas-phase acidities calculated with G1 and G2 show very good
agreement with experiment (see Table 4.2), despite the tendency for MP2 to
overestimate N-O bonds lengths.25 As expected, the G2 predictions are slightly better
than those of G1.

Table 4.2 Comparison of G1 and G2 predictions of nitromethane chemistry: heats of formation

and gas phase acidities.
∆fH0298 (kJ/mol) ∆acidH0 (kJ/mol)
G1 -87.6 1500
G2 -64.6 1486
Expt -74.8 ± 1.0 1491 ± 12

4.3.2 Complete basis set methods

The largest errors in MO thermochemical calculations result from the deficiencies of

small (finite) basis sets. With perturbative methods (MP2, MP4, etc.), it is generally
observed that successive contributions to the calculated energies decrease with the
order of the expansion, so that the sum converges to an asymptote representing the
best attainable energy for a particular basis. The convergence properties are easily
approximated if the MP2 correlation energy for a many-electron system is rewritten as

25Chemical bonds predicted by MP2 are typically too long because MP2 exaggerates the loss of
electron density within the bonding region. In an extreme case, MP2/4-31G calculations of
H2N-NO2 predict a N-N bond length of 2.106 Å, as compared to the experimental value of 1.86
Å [73].

26
 DSTO-GD-0253

a sum of electron pair energies in a particular series known as a ‘pair natural orbital’
(PNO) expansion. Complete basis set (CBS) methods use the known asymptotic
convergence of PNO expansions to extrapolate the energy of a complete basis set26
from calculations using finite basis sets: as the size of the basis set increases, the level of
applied theory decreases. Hence, CBS models typically include a HF calculation with a
very large basis set, an MP2 calculation with a smaller basis set, and one or more
higher-level calculations with a medium- to modest-sized basis.
Two typical CBS model chemistries are CBS-4 and CBS-Q. CBS-4 yields results similar
to other high-accuracy methods with substantially less computational cost; however, it
is noteworthy that the optimised geometry is obtained using a rudimentary calculation,
HF/3-21G(d). CBS-Q generally achieves better accuracies than G2, and is also less
expensive computationally. CBS-4 and CBS-Q are only capable of handling systems
composed of first- and second row atoms. More detail on these methods is given in [1].

4.3.3 Caveats for energetic materials

Care should be taken when using compound methods to model energetic molecules
containing nitro groups. The electronic structures of nitrogen oxides display strong
correlation effects because the highest-occupied and lowest-unoccupied molecular
orbitals (HOMO and LUMO) have similar energies, allowing them to ‘mix.’ Thus, an
accurate quantum chemical model of nitrogen oxides requires not only a description of
the occupied electron orbitals up to the HOMO, but also a description of the LUMO –
in technical jargon, this model is termed a ‘multiconfigurational’ treatment. Any model
of nitro compounds that relies predominantly on single-configuration methods (e.g.,
HF, MP2, MP4) is subject to error. For instance, the use of MP2 predictions with small
basis sets grossly overestimate O-N and N-N bond distances in nitrogen oxides, and
fail to predict the correct dissociation of nitromethane [74]. Such errors can corrupt the
results of highly parameterised procedures like G1, G2 and CBS-Q, which rely on
MP2/6-31G(d) geometry optimisations [1]. These errors may be avoided by the use of
other procedures such as, for instance, CBS-RAD (complete basis set approximation for
radicals), which uses a high-level configuration-interaction method, QCISD/6-31G(d),
to determine geometries and zero-point energies, followed by coupled-cluster methods
for high-accuracy calculations [75]. To date, there has been no reported use of CBS-
RAD for studies of energetic materials.

5. Conclusions
Electronic structure calculations provide useful estimates of the energetic properties of
chemical systems, including preferred molecular structures, spectroscopic features and
probable reaction paths. This review of some current electronic structure calculations

26 A complete basis set contains an arbitrary number of terms, and can therefore generate the
‘true’ molecular wavefunction.

27
DSTO-GD-0253

has concentrated on discussion of ab initio techniques, as these are the most accurate
and, in principle, ‘universal’ methods. Rapid advances in computer technology are
making computationally expensive ab initio methods increasingly more practical for
use with realistic chemical systems. In particular, ‘cheaper’ methods such as density
functional calculations and ‘layered’ models such as ONIOM are continually being
refined, and show promise of providing consistent and accurate chemical predictions
for most complex systems. Energetic materials, which often contain azide or nitro
groups that require sophisticated treatments of electronic structure, fall into this
category. A subsequent publication [4] will describe the application of electronic
structure methods to estimates of the chemical properties of energetic materials.

28
 DSTO-GD-0253

6. References
1. Foresman, J.B. and Frisch, C., (1996) Exploring Chemistry with Electronic Structure
Methods, 2nd ed., Gaussian Inc., Pittsburgh, PA.
2. Östmark, H., Langlet, A., Bergman, H., Wingborg, N., Wellmar, U., and Bemm, U.
(1998), FOX-7 – A new explosive with low sensitivity and high performance,” Proc.
11th Detonation (Int.) Symposium, Snowmass, CO.
3. Fried, L.E., Howard, W.M., and Souers, P.C. (1999), “CHEETAH 2.0 Users
Manual,” Lawrence Livermore National Laboratory.
4. Dorsett, H.E. (in preparation). A review of quantum chemical studies of energetic
materials. DSTO Technical Report.
5. Akutsu, Y. and Tahara, S.Y. (1991) Calculations of heats of formation for nitro
compounds by semi-empirical MO methods and molecular mechanics, Journal of
Energetic Materials 9 161-172.
6. Guangyao, Z. and Guanlin, H. (1991) Quantum chemical studies on structure-
property relationships of NTO, Proceedings of the 17th International Pyrotechnics
Seminar (combined with the 2nd International Beijing International Symposium on
Pyrotechnics and Explosives), Beijing Institute of Technology Press, Beijing, China,
276-278.
7. De Paz, J.L. and Ciller, J. (1993) On the use of AM1 and PM3 methods on energetic
compounds, Propellants, Explosives, Pyrotechnics 18, 33-40.
8. (a) Ritchie, J.P., and Kober, E.M. (1989) Molecular models for explosives:
applications to NTO, Proceedings of the 9th Symposium (International) on Detonation,
Portland OR, Office of the Chief of Naval Research, Arlington, VA, 1185-1192. (b)
Ritchie, J.P. (1993) The use of quantum chemistry to estimate the performance and
properties of explosives, Proceedings of the 10th Symposium (International) on
Detonation, Boston, MA, Office of the Chief of Naval Research, Arlington, VA, 157-
165.
9. Östmark, H., Bergman, H. and Aqvist, G. (1993) The chemistry of 3-Nitro-1,2,4-
triazol-5-one (NTO): thermal decomposition, Thermochimica Acta 213 165-175.
10. Bergman, H., Östmark, H., Ekvall, K. and Langlet, A. (1993) A study of the
sensitivity and decomposition of keto-RDX, Proceedings of the 10th Symposium
(International) on Detonation, Boston, MA, Office of the Chief of Naval Research,
Arlington, VA, 862-869.
11. (a) Carper, W.R., Davis, L.P., Extine, M.W. (1982) Molecular structure of 2,4,6-
Trinitrotoluene, Journal of Physical Chemistry 86, 459-462. (b) Stewart, J.J.P., Bosco,
S.R., and Carper, W.R. (1986) ‘Vibrational spectra of 2,4,6-Trinitrotoluene and its
isotopically substituted analogues,’ Spectrochimica Acta 42A, (1) 13-21. (c) Nash,
C.P., Nelson, T.E., Stewart, J.J.P., and W.R. Carper, (1989) Molecular structure and
vibrational analysis of 2,4,6-Trinitrotoluene and 2,4,6-Trinitrotoluene-α-d3,
Spectrochimica Acta 45A (5) 585-588.

29
DSTO-GD-0253

12. I.N. Levine, Quantum Chemistry, Boston: Allyn and Bacon, Inc. (1983).
13. D.B. Cook, Handbook of Computational Quantum Chemistry, New York: Oxford
University Press (1998).
14. HyperChem 5.0 user manuals, Hypercube, Inc., FL (1996).
URL - http://www.hyper.com/
15. Chem3D 3.5 user manual, CambridgeSoft, MA (1996).
URL - http://www.camsoft.com/
16. Atkins, P.W., Quanta, 2nd ed., Oxford University Press (1991).
17. (a) Sewell, T.D., and Thompson, D.L., (1991) Classical dynamics study of the
unimolecular dissociation of Hexahydro-1,3,5-trinitro-1,3,5-triazine, Journal of
Physical Chemistry 95, 6228-6242. (b) Chambers C.C., and Thompson, D.L. (1995)
Further sudies of the classical dynamics of the unimolecular dissociation of RDX,
Journal of Physical Chemistry 99 (1995) 15881-15889. (c) Shalashilin, D.V., and
Thompson, D.L. (1997) Monte Carlo variational transition state theory of the
unimolecular dissociation of RDX, Journal of Physical Chemistry A101 961-966.
18. Wallis, E.P., and Thompson, D.L. (1993) Molecular dynamics simulations of ring
inversion in RDX, Journal of Chemical Physics 99 (4) 2661-2673.
19. Rice, B.M., Grosh, J., and Thompson, D.L. (1995) Vibrational mode selectivity in the
unimolecular decomposition of CH2NNO2, Journal of Chemical Physics 102 (22)
8790-8799.
20. Kohno, Y., Ueda, K., and Imamura, A. (1996) Molecular dynamics simulations of
initial decomposition processes on the unique N-N bond in nitramines in the
crystalline state, Journal of Physical Chemistry 100, 4701-4712.
21. Sorescu, D.C., and Thompson, D.L. (1997) Crystal packing and molecular
dynamics studies of the 5-Nitro-2,4-dihydro-3H-1,2,4-triazol-3-one crystal, Journal
of Physical Chemistry B101, 3605-13.
22. (a) Sorescu, D.C., Rice, B.M., and Thompson, D.L. (1997) Intermolecular packing
for the Hexahydro-1,3,5-trinitro-1,3,5-s-triazine crystal (RDX): A crystal packing,
Monte Carlo and molecular dynamics study, Journal of Physical Chemistry B101,
798-808. (b) Sorescu, D.C., Rice, B.M., and Thompson, D.L. (1998) A transferable
intermolecular potential for nitramine crystals, Journal of Physical Chemistry A102,
8386-8392. (c) Sorescu, D.C., Rice, B.M., and Thompson, D.L. (1998) Molecular
packing and NPT-molecular dynamics investigation of the transferability of the
RDX intermolecular potential to 2,4,6,8,10,12-Hexanitrahexaazaisowurtzitane,
Journal of Physical Chemistry B102 948-952. (d) Sorescu, D.C., Rice, B.M., and
Thompson, D.L. (1998) Isothermal-isobaric molecular dynamics simulations of
1,3,5,7-Tetranitro-1,3,5,7-tetraazacyclooctane (HMX) crystals, ibid., 6692-6695.
23. (a) Sewell, T.D. (1996) Monte Carlo simulations of crystalline TATB, in
Decomposition, Combustion and Detonation Chemistry of Energetic Materials, Materials
Research Society Symposium Proceedings 418 Boston, MA (November 1995) Brill, T.B.,

30
 DSTO-GD-0253

Russell, T.P, Tao, W.C., and Wardle, R.B., Eds., Materials Research Society,
Pittsburgh, PA, 67-72. (b) Sewell, T.D. (1998) Monte Carlo and molecular dynamics
studies of crystalline RDX, HMX and TATB, in Preprints of the Proceedings of the 11th
Symposium (International) on Detonation, Snowmass, CO, Office of the Chief of
Naval Research, Arlington, VA, 115-120.
24. (a) Ritchie, J.P. (1989) Computational analysis of slip in PETN, Proceedings of the 10th
Symposium (International) on Detonation, Boston, MA, Office of the Chief of Naval
Research, Arlington, VA, 971-978. (b) Ritchie, J.P. (1993) Molecular modelling of
slip supposed to occur in the shock initiation of crystalline PETN, in Structure and
Properties of Energetic Materials, Materials Research Society Symposium Proceedings
296, Boston, MA (November, 1992) Liebenberg, D.H., Armstrong, R.W., and
Gilman, J.J., Eds., Materials Research Society, Pittsburgh, PA, 99-106.
25. (a) Kunz, A.B. (1994) A calculation of the cohesive properties of solid methane by a
local orbital approach, including correlation, Journal of Physics, Condensed Matter, 6
(17) L233-L237. (b) Kunz, A.B., (1996) Ab initio investigation of the structure and
electronic properties of the energetic solids TATB and RDX, Physical Review B53
(15) 9733-9738. (c) Kunz, A.B. (1996) An ab initio investigation of crystalline PETN,
in Decomposition, Combustion and Detonation Chemistry of Energetic Materials,
Materials Research Society Symposium Proceedings 418, Boston, MA (November 1995)
Brill, T.B., Russell, T.P, Tao, W.C., and Wardle, R.B., Eds., Materials Research
Society, Pittsburgh, PA, 287-292.
26. Dzyabchenko, A.V., Pivina, T.S., and Arnautova, E.A. (1996) Prediction of
structure and density for organic nitramines, Journal of Molecular Structure 378, 67-
82.
27. Filippini, G., and Gavezzotti, A., (1994) The crystal structure of 1,3,5-Triamino-
2,4,6-trinitrobenzene. Centrosymmetric or non-centrosymmetric? Chemical Physics
Letters 231, 86-92.
28. (a) Phillips, L., Sinkovits, R.S., Oran, E.S., and Boris, J.P., (1993) The interaction of
shocks and defects in Lennard-Jones crystals, Journal of Physics, Condensed Matter 5,
6357-6376. (b) Phillips, L. (1995) Initiation of detonations in three-dimensional
crystals with defects, Journal of Physics, Condensed Matter 7, 7813-7822.
29. (a) Tsai, D.H. (1993) Hot spot heating from impurities and vacancies in a
crystalline solid under rapid compression, in Structure and Properties of Energetic
Materials, Materials Research Society Symposium Proceedings 296, Boston, MA
(November, 1992) Liebenberg, D.H., Armstrong, R.W., and Gilman, J.J., Eds.,
Materials Research Society, Pittsburgh, PA, 113-120. (b) Tsai, D.H. (1996) Hot spots
in a molecular solid under rapid compression: energy sharing among the T-R-V
degrees of freedom, in Decomposition, Combustion and Detonation Chemistry of
Energetic Materials, Materials Research Society Symposium Proceedings 418, Boston,
MA (November 1995) Brill, T.B., Russell, T.P, Tao, W.C., and Wardle, R.B., Eds.,
Materials Research Society, Pittsburgh, PA, 281-286. (c) Tsai, D.H. (1991) Structural
defects and 'hot spot' formation in a crystalline solid under rapid compression. I.
Vacancy clusters and slip bands, Journal of Chemical Physics 95 (15) 7497-7503. (d)

31
DSTO-GD-0253

Tsai, D.H., and Trevino, S.F. (1984) Simulation of the initiation of detonation in an
energetic molecular crystal, Journal of Chemical Physics 81 (12) 5636-5637. (e) Tsai,
D.H., and Armstrong, R.W. (1994) Defect-enhanced structural relaxation
mechanism for the evolution of hot spots in rapidly compressed crystals, Journal of
Physical Chemistry 98 (43) L10997-L11000.
30. Haskins, P.J., and Cook, M.D. (1998) Molecular dynamics studies of fast
decomposition in energetic molecules, in Preprints of Proceedings of the 11th
Symposium (International) on Detonation, Snowmass, CO, Office of the Chief of
Naval Research, Arlington, VA, 104-108, and references therein.
31. Elert, M.L., Robertson, D.H., and White, C.T (1996) Molecular dynamics study of
the effect of varying exothermicity on the properties of condensed-phase
detonation, in Decomposition, Combustion and Detonation Chemistry of Energetic
Materials, Materials Research Society Symposium Proceedings 418 Boston, MA
(November 1995) Brill, T.B., Russell, T.P, Tao, W.C., and Wardle, R.B., Eds.,
Materials Research Society, Pittsburgh, PA, 309-312.
32. Barrett, J.C. Barrett, Robertson, D.H., Brenner, D.W., and White, C.T. (1996)
Simulations of ozone detonation using a reactive empirical bond order (REBO)
potential for the oxygen system, in Decomposition, Combustion and Detonation
Chemistry of Energetic Materials, Materials Research Society Symposium Proceedings 418
Boston, MA (November 1995) Brill, T.B., Russell, T.P, Tao, W.C., and Wardle, R.B.,
Eds., Materials Research Society, Pittsburgh, PA, 301-308.
33. Soulard, L. (1996) Molecular dynamics calculations on the properties of the
reaction zone in liquid explosive, in Decomposition, Combustion and Detonation
Chemistry of Energetic Materials, Materials Research Society Symposium Proceedings 418
Boston, MA (November 1995) Brill, T.B., Russell, T.P, Tao, W.C., and Wardle, R.B.,
Eds., Materials Research Society, Pittsburgh, PA, 293-300.
34. URL –http://www.amber.ucsf.edu/amber/amber.html
35. URL - http://www.lobos.nih.gov/Charmm
36. Rapaport, D.C. (1995) The Art of Molecular Dynamics Simulation, Cambridge
University Press, Melbourne, Australia.
37. Haile, J.M. (1992) Molecular Dynamics Simulation: Elementary Methods, John Wiley
and Sons Inc., Brisbane, Australia.
38. Rappé, A.K. and Casewit, C.J. (1997) Molecular Mechanics Across Chemistry,
University Science Books, Sausalito, CA.
URL - http://www.chm.colostate.edu/mmac/
39. Matsubara, T., Sieber, S., and Morokuma, K. (1996) A test of the new integrated
MO + MM (IMOMM) method for the conformational energy of ethane and
n-butane, International Journal of Quantum Chemistry 60, 1101-1109.

32
 DSTO-GD-0253

40. Musaev, D.G., Froese, R.D.J., and Morokuma, K. (1998) Molecular orbital and
IMOMM studies of the chain transfer mechanisms of the diimine-M(II), (M=Ni,
Pd) catalyzed ethylene polymerization reaction, Organometallics 17, 1850-1860.
41. Froese, R.D.J., Musaev, D.G., and Morokuma, K. (1998) A theoretical study of
substituent effects in the diimine-M(II) catalyzed ethylene polymerization reaction
using the IMOMM method” Journal of the American Chemical Society 120, 1581-1587.
42. Humbel, S., Sieber, S., and Morokuma, K. (1996) The IMOMO method: integration
of different levels of molecular orbital approximations for geometry optimization
of large systems. Test for n-butane conformation and SN2 reaction: RCl +Cl-,
Journal of Chemical Physics 105, 1959-1967.
43. Svensson, M., Humbel, S., and Morokuma, K. (1996) Energetics using the single
point IMOMO (Integrated Molecular Orbital + Molecular Orbital) calculations.
Choices of computational levels and model systems, Journal of Chemical Physics 105,
3654-3661.
44. Froese, R.D.J., and Morokuma, K. (1999) IMOMO-G2MS approaches to accurate
calculations of bond dissociation energies of large molecules, Journal of Physical
Chemistry A103, 4580-4586.
45. Svensson, M., Humbel, S., Froese, R.D.J., Matsubara, T., Sieber, S., and Morokuma,
K. (1996) ONIOM: A multilayered integrated MO+MM method for geometry
optimizations and single point energy predictions. A test for Diels-Alder reactions
and Pt(P(t-Bu)3)2 + H2 oxidative addition, Journal of Physical Chemistry 100, 19357–
19363.
46. Froese, R.D.J., and Morokuma, K. (1999) Accurate calculations of bond breaking
energies in C60 using the three-layered ONIOM method” Chemical Physics Letters
305, 419-424.
47. Atkins, P.W. and Friedman, R.S. (1997) Molecular Quantum Mechanics, 3rd ed.,
Oxford University Press Inc., New York, NY.
48. Hehre, W.J., Radom, L., Schleyer, P.v.R. and Pople, J.A. (1986) Ab initio Molecular
Orbital Theory, Wiley-Interscience, Brisbane, Australia.
49. Simons, J., and Nichols, J. (1997) Quantum Mechanics in Chemistry, Oxford
University Press, Oxford, UK.
50. Frisch, M.J., Trucks, G.W., Schlegel, H.B., Scuseria, G.E., Robb, M.A., Cheeseman,
J.R., Zakrzewski, V.G., Montgomery, J.A. Jr., Stratmann, R.E., Burant, J.C.,
Dapprich, S., Millam, J.M., Daniels, A.D., Kudin, K.N., Strain, M.C., Farkas, O.,
Tomasi, J., Barone, V., Cossi, M., Cammi, R., Mennucci, B., Pomelli, C., Adamo, C.,
Clifford, S., Ochterski, J., Petersson, G.A., Ayala, P.Y., Cui, Q., Morokuma, K.,
Malick, D.K., Rabuck, A.D., Raghavachari, K., Foresman, J.B., Cioslowski, J., Ortiz,
J.V., Stefanov, B.B., Liu, G., Liashenko, A., Piskorz, P., Komaromi, I., Gomperts, R.,
Martin, R.L., Fox, D.J., Keith, T., Al-Laham, M.A., Peng, C.Y., Nanayakkara, A.,
Gonzalez, C., Challacombe, M., Gill, P.M.W., Johnson, B., Chen, W., Wong, M.W.,

33
DSTO-GD-0253

Andres, J.L., Gonzalez, C., Head-Gordon, M., Replogle, E.S., and Pople, J.A. (1998)
Gaussian 98, Rev. A.3, Gaussian Inc., Pittsburgh PA.
51. Frisch, A. and Frisch, M.J. (1998) Gaussian 98 User’s Reference, Gaussian Inc.,
Pittsburgh, PA.
52. D. Habibollahzadeh, M. Grodzicki, J.M. Seminario and P. Politzer, “Computational
Study of the Concerted Gas-Phase Triple Dissociations of 1,3,5-Triazacyclohexane
and its 1,3,5-Trinitro Derivative (RDX),” J. Phys. Chem. 95 (1991) 7699-7702.
53. B.M. Rice and C.F. Chabalowski, “Ab Initio and Nonlocal Density Functional Study
of 1,3,5-trinitro-s-triazine (RDX) Conformers,” J. Phys. Chem. A101 (1997) 8720-
8726.
54. Parr, R.G. and Yang, W. (1989) Density-Functional Theory of Atoms and Molecules, in
International Series of Monographs on Chemistry, Breslow, R., Goodenough, J.B.,
Halpern, J., and Rolinson, J., Eds., Oxford University Press, New York.
55. Joubert, D., Ed. (1998) Density Functionals: Theory and Applications, Springer-Verlag,
Heidelberg, DE.
56. Lee, C., Yang W., and Parr, R.G. (1988) Development of the Colle-Salvetti
correlation energy formula into a functional of the electron density, Physical
Review B37 (2 ) 785-789.
57. Kohn, W., Becke A.D., and Parr, R.G. (1996) Density functional theory of electronic
structure, Journal of Physical Chemistry 100 (31) 12974-12980.
58. Durant, J.L. (1996) Evaluation of transition state properties by density functional
theory, Chemical Physics Letters 256, 595-602.
59. (a) Becke, A.D. (1998) Density functional exchange energy approximation with
correct asymptotic behaviour, Physical Review A38 (6) 3098-3100. (b) Becke, A.D.
(1993) Density functional thermochemistry. III. The role of exact exchange, Journal
of Chemical Physics 98 (7) 5648-5652.
60. Perdew, J.P., Burke, K., and Wang, Y. (1996) Generalised gradient approximation
for the exchange-correlation hole of a many-electron system, Physical Review B54
(23) 16533-16539.
61. Pople, J.A., Krishnan, R., Schlegel, H.B., and Binkley, J.S. (1978) Electron
correlation theories and their application to the study of simple reaction potential
surfaces, International Journal of Quantum Chemistry 14, 545-560.
62. Farnell, L., Pople, J.A., and Radom, L. (1983) Structural predictions for open-shell
systems: a comparative assessment of ab initio procedures, Journal of Physical
Chemistry 87, 79-82.
63. Jensen, F. (1990) The remarkably large effect of spin contamination on calculated
vibrational frequencies, Chemical Physics Letters 169 (6) 519-528.

34
 DSTO-GD-0253

64. Schegel, H.B. (1986) Potential energy curves using unrestricted Møller-Plesset
perturbation theory with spin annihilation, Journal of Chemical Physics 84 (8) 4530-
4534.
65. Martin, J.M.L., François, J.P., and Gijbels, R. (1990) Unusually large effect of single
excitations on the geometry of radical species and limiting spin-projection
invariance of some correlated methods, Chemical Physics Letters 166 (3) 295-302.
66. Baker, J., Scheiner, A., and Andzelm, J. (1993) Spin Contamination in Density
Functional Theory, Chemical Physics Letters 216 (3-6) 380-388.
67. Dunning, T.H. (1989) Gaussian basis sets for use in correlated molecular
calculations. I. The atoms boron through neon and hydrogen, Journal of Chemical
Physics 90 (2) 1007-1023.
68. Pople, J.A., Head-Gordon, M., Fox, D.J., Raghavachari, K., and Curtiss, L.A. (1989)
Gaussian-1 theory: a general method for prediction of molecular energies, Journal
of Chemical Physics 90, 5622.
69. (a) Curtiss, L.A., Raghavachari, K., Trucks, G.W., and Pople, J.A. (1991) Gaussian-2
theory for molecular energies of first and second row compounds, Journal of
Chemical Physics 94, 7221. (b) Mebel, A.M., Morokuma, K., and Lin, M.C. (1995)
Modification of the Gaussian-2 theoretical model: the use of coupled-cluster
energies, density-functional geometries and frequencies, Journal of Chemical Physics
103 (17) 7414-7421.
70. Lammertsma, K. and Prasad, B. (1993) “Nitro ↔ aci-nitro tautomerism,” J. Am.
Chem. Soc. 115, 2348-2351.
71. Mayer, P. and Radom, L. (1998) “Deprotonating molecules and free radicals to
form carbon-centered anions: A G2 study of molecular and free radical acidity,” J.
Phys. Chem. A102, 4918-4924.
72. Seminario, J . and Politzer, P. (1992) “G2 and density functional studies of H2N-
NO2 dissociation, inversion and isomerisation,” Int. J. Quan. Chem. Symp. 26, 497-
504.
73. Marynick, D.S., Ray A.K., and Fry, J.L (1985) Electronic structure calculations on R-
NO2 species, R = H, CH3, NH2, HO and F,” Chemical Physics Letters 116, 429-433.
74. (a) McKee, M.L. (1989) MCSCF study of the rearrangement of nitromethane to
methyl nitrite, Journal of Physical Chemistry 93, 7365-7369. (b) McKee, M.L. (1986) Ab
initio study of rearrangements on the CH3NO2 potential energy surface, Journal of
the American Chemical Society 108, 5784-5792. (c) McKee, M.L. (1984) A study of the
CH2NO2 radical using a multiconfiguration self-consistent field approach,” Journal
of Chemical Physics 81 (8) 3580-3587.
75. Mayer, P.M., Parkinson, C.J., Smith, D.M, and Radom, L. (1998) An assessment of
theoretical procedures for the calculation of reliable free radical thermochemistry:
a recommended new procedure,” Journal of Chemical Physics 108 (2) 604-615.

35
DSTO-GD-0253

36
 DISTRIBUTION LIST

Overview of Molecular Modelling and Ab initio Molecular Orbital Methods

Suitable for Use with Energetic Materials

H.Dorsett and A. White

AUSTRALIA

DEFENCE ORGANISATION

S&T Program
Chief Defence Scientist 
FAS Science Policy  shared copy
AS Science Corporate Management 
Director General Science Policy Development
Counsellor Defence Science, London (Doc Data Sheet )
Counsellor Defence Science, Washington (Doc Data Sheet )
Scientific Adviser to MRDC Thailand (Doc Data Sheet )
Scientific Adviser Policy and Command
Navy Scientific Adviser (Doc Data Sheet and distribution list only)
Scientific Adviser - Army (Doc Data Sheet and distribution list only)
Air Force Scientific Adviser
Director Trials

Aeronautical and Maritime Research Laboratory

Director
Chief of Weapons Systems Division
Research Leader Joint and Intelligence
Head Explosives
Head Propulsion Systems Technology
Head Pyrotechnics
Head Terminal Effects
Author(s):
Dr. A. White (5 copies)
Dr. H. Dorsett (5 copies)

DSTO Library and Archives

Library Fishermans Bend (Doc Data sheet only)
Library Maribyrnong (Doc Data sheet only)
Library Salisbury (2 copies)
Australian Archives
Library, MOD, Pyrmont (Doc Data sheet only)
US Defense Technical Information Center, 2 copies
UK Defence Research Information Centre, 2 copies
Canada Defence Scientific Information Service, 1 copy
NZ Defence Information Centre, 1 copy
National Library of Australia, 1 copy
Capability Development Division
Director General Maritime Development (Doc Data Sheet only)
Director General Land Development (Doc Data Sheet only)
Director General C3I Development (Doc Data Sheet only)
Director General Aerospace Development (Doc Data Sheet only)

Army
ABCA Standardisation Officer, Puckapunyal, (4 copies)
SO (Science), DJFHQ(L), MILPO Enoggera, Queensland 4051 (Doc Data Sheet
only)
NAPOC QWG Engineer NBCD c/- DENGRS-A, HQ Engineer Centre Liverpool
Military Area, NSW 2174 (Doc Data Sheet only)
Intelligence Program
DGSTA Defence Intelligence Organisation
Defence Intelligence Organisation-Information Centre
Corporate Support Program
Library Manager, DLS-Canberra (Doc Data sheet only)

UNIVERSITIES AND COLLEGES

Australian Defence Force Academy
Library
Head of Aerospace and Mechanical Engineering
Dr. Kenneth R. Harris
Senior Librarian, Hargrave Library, Monash University
Librarian, Flinders University

OTHER ORGANISATIONS

NASA (Canberra)
AusInfo

OUTSIDE AUSTRALIA

ABSTRACTING AND INFORMATION ORGANISATIONS

Library, Chemical Abstracts Reference Service
Engineering Societies Library, US
Materials Information, Cambridge Scientific Abstracts, US
Documents Librarian, The Center for Research Libraries, US

INFORMATION EXCHANGE AGREEMENT PARTNERS

Acquisitions Unit, Science Reference and Information Service, UK
Library - Exchange Desk, National Institute of Standards and Technology, US

SPARES (5 copies)

Total number of copies: 56

 Page classification: UNCLASSIFIED

DEFENCE SCIENCE AND TECHNOLOGY ORGANISATION

DOCUMENT CONTROL DATA 1. PRIVACY MARKING/CAVEAT (OF
DOCUMENT)

2. TITLE 3. SECURITY CLASSIFICATION (FOR UNCLASSIFIED

REPORTS THAT ARE LIMITED RELEASE USE (L) NEXT
Overview of Molecular Modelling and Ab initio Molecular Orbital Methods TO DOCUMENT CLASSIFICATION)
Suitable for Use with Energetic Materials
Document (U)
Title (U)
Abstract (U)

4. AUTHOR(S) 5. CORPORATE AUTHOR

H. Dorsett and A. White Aeronautical and Maritime Research Laboratory

PO Box 1500
Salisbury SA 5108 Australia

6a. DSTO NUMBER 6b. AR NUMBER 6c. TYPE OF REPORT 7. DOCUMENT

DSTO-GD-0253 AR-011-578 General Document DATE
September 2000
8. FILE NUMBER 9. TASK NUMBER 10. TASK SPONSOR 11. NO. OF PAGES 12. NO. OF
9505/19/05/01 DST 98/077 DSTO 36 REFERENCES
75
13. URL on the World Wide Web 14. RELEASE AUTHORITY

http://www.dsto.defence.gov.au/corporate/reports/DSTO-GD-0253.pdf Chief, Weapons Systems Division

15. SECONDARY RELEASE STATEMENT OF THIS DOCUMENT

Approved for public release

OVERSEAS ENQUIRIES OUTSIDE STATED LIMITATIONS SHOULD BE REFERRED THROUGH DOCUMENT EXCHANGE, PO BOX 1500, SALISBURY, SA 5108
16. DELIBERATE ANNOUNCEMENT

No Limitations

17. CASUAL ANNOUNCEMENT Yes

18. DEFTEST DESCRIPTORS

computational chemistry, energetic materials, quantum theory, molecular modelling

19. ABSTRACT
This is a review of molecular modelling techniques which may be applied to studies of energetic materials. It
focusses on ab initio ('first-principles') molecular orbital calculations, since these methods offer the greatest
accuracy. Since ab initio calculations are very computer-intensive, approximate MO methods are also
discussed, which offer reasonably accurate predictions with reduced calculation times. These approximate
methods include density functional theory and 'layered' techniques which combine different levels of
theoretical sophistication into one calculation.

Page classification: UNCLASSIFIED