Bordoni PH D

Università degli Studi di Milano
Facoltà di Scienze Matematiche, Fisiche e Naturali

Dottorato di Ricerca in Fisica, Astrofisica e Fisica Applicata
Systematic calculation of molecular vibrational

spectra through a complete Morse expansion
.
Tesi di Dottorato di Andrea Bordoni PhD Thesis of

Tutore Nicola Manini Tutor
Cotutore Stefano Oss Co-tutor
Revisore Per Jensen Referee
Coordinatore Gianpaolo Bellini Coordinator
Dottorato di Ricerca XIX ciclo, anno 2005-2006

Matricola R05433
Settore scientifico-disciplinare FIS/03
Contents
Chapter 1. Introduction 5
1.1. A simple example 8
Chapter 2. The Morse oscillator: bases and matrix elements 11
2.1. The Schödinger equation for the Morse potential 11
2.2. Molecular potential approximation 14
2.3. Expansion in the Morse basis 16
2.4. Beyond the Morse basis 24
2.5. Morse and SUSY Shape Invariant Potentials 24
Chapter 3. The potential expansion 43
3.1. Potential fitting schemes 45
Chapter 4. Applications to diatomic systems 51
4.1. The Lennard-Jones potential 51
4.2. The H2 molecule 53
Chapter 5. Unlinking of basis and potential parameters 57
5.1. The algebraic form of the general Morse Hamiltonian 57
5.2. Hamiltonian and basis parameters 59
5.3. Different sB and sP 61
5.4. Bases and efficiency 62
Chapter 6. Interacting oscillators 71
6.1. Multidimensional potential Morse expansion 72
6.2. The HCAO model for triatomic molecules 73
6.3. Comparisons 76
Chapter 7. Discussion and outlook 81
Chapter 8. Programs and codes 85
Chapter 9. Appendixes 87
9.1. Appendix A: Laguerre polynomial integrals 87
9.2. Appendix B: Morse 1D and 2D 89
9.3. Appendix C: Morse expansion derivatives 90
9.4. Appendix D: Matrix elements in the QNSB 90
9.5. Appendix E: General properties of matrix elements 93
9.6. Appendix F : Dynamical Symmetry approach 93
Bibliography 99
3
CHAPTER 1
Introduction
The ever improving accuracy of spectroscopic experiments is a major stimulus to the de-
velopment of new theoretical methods and to the refining of older ones. Nowadays, the exper-
imental resolution in infrared (IR) (10−3 cm−1 ) and MW (10−7 cm−1 ) spectroscopy is a great
challenge for theoretical physicists and chemists, both engaged in exploiting their theoretical
tools to be able to understand and possibly predict what experiments detect. A variety of
methods was developed and is presently employed, based on very different approaches. In our
approach we refer to the commonly employed Born-Oppenehimer approximation: electronic
and nuclear motion decouple, and the nuclear motion is controlled by an effective potential
energy surface (PES), which we assume can be inferred from experimental data analysis or
from ab initio calculations. For these attempts to be successful an accurate determination of
the Hamiltonian of the specific system is necessary, and this can hardly be considered a solved
problem. The Hamiltonian description requires the knowledge of the adiabatic molecular PES
which can be determined numerically at the price of computationally intensive ab initio calcula-
tions, and whose specific functional form depends crucially on the coordinates chosen. Likewise,
the kinetic operator can assume different forms depending on the coordinates, and its complex-
ity can be an obstacle in real applications. Most methods try to find a compromise between
these aspects, but they often suffer in generality. There are several specialized methods suitable
only to treat molecules of particular geometry; for example, usually models aimed at treating
linear molecules are not good for nonlinear ones [1] (for a review, see Ref. [2]).
Among the various methods available, those based on normal-mode coordinates, starting
from harmonic approximation of molecular vibrations, deserve a special mention, for historical
reasons, and for the success they achieved. These methods were and still are employed largely,
from the simple formulation to more sophisticated modifications.
In its simplest form, the traditional vibrational Hamiltonian in dimensionless normal coor-
dinates Qi is assumed of the form
1X X X
(1) Ĥ = ωi (P̂i2 + Q̂2i ) + Φijk Qi Qj Qk + Φijkl Qi Qj Qk Ql + ... ,
2 i
ijk ijk
where the zero-order harmonic oscillator Hamiltonian is evident, and the PES is expanded
in (truncated) power series of the normal coordinates. ωi are the frequencies of the normal
modes, and Φijk... the force constants for the specific molecule. The eigenfunctions of the full
Hamiltonian are expanded in the complete set of harmonic-oscillator eigenfunctions, and the
quantum-mechanical dynamics is then treated perturbatively or variationally. The harmonic-
oscillator expansion gives some advantages:
• formal simplicity of the Hamiltonian;
• the potential is a simple Taylor expansion around a minimum;
• the matrix elements of power terms are exactly and algebraically computable, once
the customary harmonic-oscillator ladder operators â and aˆ† are introduced, and they
produce sparse matrices.
On the other hand, this approach has several known problems:
5
6 1. INTRODUCTION
Energy / depth of the well

1 actual APES
unphysical
leakage unbounded
region
tunneling
Es
0 PP
0 qs 0.5
q
Figure 1.1. 1D-molecular potential, and a pictorial view of a Taylor potential

expansion truncated to an odd power: unphysical tunneling may emerge, though
the potential is locally well approximated. In more than one dimension, the
problem is even more serious.
• the potential expansion in normal-mode coordinates is intrinsically local: it works well

in a small neighborhood of a minimum, but it always diverges to ∞, and also possibly
to −∞, for large positive or negative Qi values. In a polyatomic context, such a
behavior often occurs also in energy relevant regions, as pointed out for example in
Ref. [3], and sketched in Fig. 1.1).
• the convergence of the potential expansion is slow. It was recognized (see for example
Ref. [4]) that expansions of the potential as [5]
X ∆R n X ∆R n
(2) V (R) = cn or V (R) = cn ,
n
R e n
R
where ∆R is the displacement from equilibrium position Re , both give more compact
expansions.
• as the expansion diverges, near-dissociation phenomena are not described well;
• normal coordinates are particularly suitable to deal with collective vibrations, and by
definition partly describe also interactions among different bond vibrations. They are
not suitable to describe high energy excited states [2, 6] (close to dissociation, see
Chapter 6), where energy localization on individual local vibrations takes place.
As alternative (or complementary) to normal-mode-based models, local-mode models were
proposed. They were already known since early works on molecular vibrations [7, 8], but the
simplicity of the harmonic-oscillator-based models was largely preferred. An accurate descrip-
tion of the high energy excitation presently explored in experiments favoured a renaissance of
these methods [1, 9, 10, 11, 12, 13, 14]. These are based on a description in terms of suit-
able internal coordinates, reflecting better the physics of the individual bonds. This particular
choice is inspired by the necessity of the different models employed, and the most popular are
probably, beside those of Eq. (2), Morse coordinate y = 1 − e−α(R−Re ) , and y = 2(R−R R+Re
e)
[15].
The potential is usually written as a power expansion in suitable internal variables, thus its form
is particularly simple. Unfortunately, the kinetic term can assume very involved forms when
expressed in terms of internal coordinates. A method to obtain straightforwardly the kinetic
energy operator in any internal coordinate system was introduced in Ref. [16]. Nonetheless
1. INTRODUCTION 7
due to needs of manageability, the complete correct kinetic expression is seldom used. To
summarize, internal coordinate expansions have several advantages:
• the potential expansion in internal coordinate is rapidly converging;
• the convergence range is wider: the approximation is not purely local;
• the near-dissociation energy region is usually approximated well.
On the other hand, there are also some common drawbacks:
(1) spurious states appear in the form of redundant coordinates, whose elimination is a
difficult task [17];
(2) the kinetic operator assumes an intricate form;
(3) evaluation of matrix elements in the relevant basis are usually nontrivial, as no simple
algebraic relation apply as in the case of harmonic oscillator.
Variational methods are generally the best suited in quantum-mechanical problems such as
that of molecular vibrations: the matrix representation of the full Hamiltonian in some suitable
basis must be diagonalized. Exactness, simplicity and if possible sparseness of the matrix are
all basic ingredients.
In practical applications, the matrix size also plays a relevant role. Even with today’s fast
and powerful computers, rather strict constraints on the size of computationally manageable
matrices apply. The size is determined by the interplay of different factors:
• number of constituents atoms, i.e. dimensionality of the underlying quantum mechan-
ical dynamics;
• efficiency of the coordinate system employed in approximating the potential;
• efficiency of the coordinate-system-related basis employed in giving simple/sparse ma-
trix representation of the Hamiltonian.
Several different and sophisticated methods have been devised to find efficient representa-
tions aimed at reducing computational efforts, such the discrete-variable representation [18].
The present work attempts to contribute to this subject analyzing in detail a particular choice
of internal coordinate, the Morse coordinate, that we indicate as
(3) v(x) = e−α(x−x0 ) − 1 ,
where x is a suitable local mode coordinate, associated with a potential energy minimum, and
α is a free parameter of dimension inverse length. We will show that it is possible to combine
the ability of a local-mode expansion in approximating efficiently a wide class of vibrational
potentials, with some advantages of a harmonic expansion, namely basis completeness, the
simplicity and exactness of matrix element evaluation, and the sparseness of the resulting
matrix. We develop and analyze the model from scratch, starting from the simple diatomic
case, where the model can be explored exactly and without approximations, and compared to
simple numerical exact solutions.
In fact, diatomic molecules are not completely trivial systems at all. They still represent an
interesting ground on their own (see for example Refs. [15, 19, 20, 21, 22]). The very high
accuracy of spectroscopic data available for these simple molecules challenges the accuracy of
theoretical models, and makes the need of inclusion of carefully modelled anharmonicity effects
as well as non-adiabatic contributions a must. Diatomics still constitute a significant testing
ground for techniques aimed at being employed in polyatomic systems.
The plan of the present work is the following. Chapter 2 introduces our model in all its
technical details, showing the similarity, differences and advantages with respect to a traditional
Morse basis expansion. Chapter 3 concerns the problem of the approximation of a general 1D
potential by a power expansion in v(x), with particular attention to the compromise between
local accuracy and global behavior. In Chap. 4 the method applies to a couple of realistic
8 1. INTRODUCTION
B
0
A
Taylor expansion (10)
Taylor epansion (4)
Energy [ALJ/4]
Morse
-0.5 LJ
-1
1 2
1/6
x [2 Σ]
Figure 1.2. Approximations of the LJ potential (solid): the Morse potential

(dashed), with equal position of the minimum, second derivative in the mini-
mum, and well depth; the 4th order (dot-dashed) and 10th order (dotted) Taylor
expansion of VLJ . The points A and B are characterized by V (x) = 1% V (x0 ),
i.e. 99% of the dissociation energy from the well bottom.
potentials: the analytic Lennard-Jones potential and the ab-initio H2 potential as determined
in Ref [23]. Chapter 5 is devoted to a generalization of the method, with a focus on optimiza-
tion: some remarks presented could be useful to tune the parameters in applications where
computational costs need to be kept at the minimum. Chapter 6 collects preliminary work on
the extension to polyatomic systems. Chapter 8 summarizes the codes developed to implement
the several tasks involved in the formalism.
1.1. A simple example

To introduce the main discussion, we illustrate some of the above observation with an
example. We apply the harmonic-oscillator-based method to the Lennard-Jones (LJ) potential
that will be extensively studied in the following.
Figure 1.2 compares the LJ potential (for details refer to Chaps. 3 and 4 below), together
with its Taylor expansions around its minimum, truncated to order 4 and 10. Even highest
order assures that the 1D expansion does not diverge to −∞, thus keeping the problem well
defined. However, the +∞ divergence at large x is unavoidable. Note that:
• the dissociation region is not described at all, even for the 10th order expansion, and
agreement does not improve for higher order expansions;
• the improvement of the accuracy is slow, and limited to a local region around the
minimum;
• the pure Morse potential of equal minimum position and curvature, and equal well
depth, gives a better global accuracy, even though agreement is obviously far from
perfect.
For the same problem, Table 1.1 compares the eigenvalues obtained for the two Taylor
expansions and the Morse expansion. As expected, the agreement improves with the degree
of expansion but very slowly and only for few low-energy states. An accurate description
of the low-lying levels requires a high-order Taylor expansion, but the dissociation region is
described poorly anyway. By comparison, the spectrum of the Morse oscillator, adapted to the
1.1. A SIMPLE EXAMPLE 9
n LJ Taylor 4 Taylor 4 Taylor 10 Pure Morse

(exact) (exact) (pert. theory) (exact) (analytical)
0 -0.8824 -0.8791 (0.3%) -0.8842 (-0.2%) -0.8824 (0.001%) -0.8818 (0.06%)
1 -0.6749 -0.6419 (3%) -0.6899 (-1.5%) -0.6742 (0.07%) -0.6677 (0.7%)
2 -0.5014 -0.3868 (11%) -0.5588 (-6%) -0.4913 (1%) -0.4833 (1.8%)
3 -0.3595 -0.1094 (25%) -0.5079 (-15%) -0.3065 (5%) -0.32862 (3%)
4 -0.2464 (positive -0.5542 (-30%) -0.0989 (15%) -0.20369 (4%)
5 -0.1593 energy) -0.7149 (-55%) (positive -0.10850 (5%)
6 -0.0951 energy) -0.04304 (5%)
7 -0.0508 -0.00732 (4%)
8 -0.0228 (missing)
9 -0.0075 (missing)
10 -0.0013 (missing)
Table 1.1. Comparison of the exact eigenvalues of the LJ potential of Fig. 1.2
with those of the 4th order traditional Taylor expansion (dot-dashed curve in
Fig. 1.2) obtained by exact (finite differences) integration of the differential
problem, and by second order perturbation theory. The eigenvalues of the 10th
order polynomial well (dotted curve in Fig. 1.2) and those of the Morse potential
(dashed curve in Fig. 1.2) are also reported for comparison. The differences in
brackets are in percent of the well depth.
LJ potential, works substantially better: even though few bound eigenstates near dissociation
are missing, it is clear that the Morse well realizes a far better starting point to describe the
asymmetric well. The addition of some comparably small corrections to the Morse potential
should make the well wider near dissociation (Fig. 1.2), thus retrieving a quantitatively accurate
spectrum. The present work addresses precisely the problem of obtaining a quantitatively good
approximation of an arbitrary molecular potential, and of solving efficiently the associated
quantum-mechanical problem.
CHAPTER 2
The Morse oscillator: bases and matrix elements
2.1. The Schödinger equation for the Morse potential

The Morse potential was introduced [24] with the purpose of producing an accurate de-
scription of typical spectra of diatomic molecules through an analytically solvable Schrödinger
problem. We set our study within the Born-Oppenheimer scheme, the motion being driven by
the electronic ground-state PES, and assume that a wide electronic gap prevents any vibration-
induced transition to the excited electronic potential-energy surfaces. We therefore consider
the 3D Schödinger equation for the adiabatic motion of the internal coordinates of a diatomic
molecule:
~2 2
(4) − ∇ Ψ(x) + V (x)Ψ(x) = W Ψ(x) ,
2µ
where µ = MM11+M M2
2
is the reduced mass expressed in terms of the masses M1 , M2 of the atomic
constituents. x stands for a 3-dimensional coordinate, which can be expressed in polar coordi-
nates (r, θ, φ). The adiabatic potential V (x) = V (r) is the sum of the intramolecular Coulomb
2
repulsion e Zr1 Z2 plus the total electronic energy relative to that given fixed nuclear distance.
Through some standard manipulation, Eq. (4) can be cast in a more convenient form. If
we assume that the wavefunction takes the (standard) factorized form
R(r)
(5) Ψ = N Yjm (θ, φ) ,
r
we obtain the following equation for the radial component R(r):
d2 R j(j + 1)R 2µ
(6) + + 2 [W − V (r)]R = 0 .
dr 2 r2 ~
The exact form of the potential energy V (r) could be determined, for example, by accurate
electronic-structure methods. In the present work we will approximate the exact V (x) with a
suitable expansion, starting with the Morse potential:
(7) VM (r) = V0 e−2α(r−r0 ) − 2V0 e−α(r−r0 ) .

As shown in Fig. 2.1, the Morse potential represents an asymmetric potential well, an in-
trinsically anharmonic oscillator, with a dissociation threshold located at energy V0 above its
minimum.
The main features of VM are: its single well defined minimum, a finite dissociation energy
obtained for large r, a strongly repulsive region at r r0 , a finite number L of bound states
and a continuum of scattering states representing free motion in the unbound region r r0 .
These are indeed all the features of a typical diatomic molecular (model) potential, and so
Morse potential is expected to be a fair starting point. The three quantities V0 (depth of the
well), α (shape parameter), and r0 (position of the minimum) entering the definition (7) allow
some degree of tuning, making a Morse choice sufficiently flexible.
11
12 2. THE MORSE OSCILLATOR: BASES AND MATRIX ELEMENTS
V(r) 0
V0
PSfrag replacements V0 (1 − e−1 )2
0 r0 1 r
α
Figure 2.1. Morse potential general shape. The role of the three parameters
entering the definition (7) is illustrated.
Once V (r) is replaced by VM (r), Eq. (6) is further manipulated, and for j = 0 (rotational
ground state), after the coordinate change u = r − r0 , becomes
d2 R 2µ
(8) + 2 [W − V0 e−2αu + 2 V0 e−αu ]R = 0 .
du2 ~
R(u) must be continuous, finite and single valued in the range −∞ ≤ u ≤ ∞; |R(u)| → 0 for
u → ±∞. By applying the substitution y = e−αu , we rewrite Eq. 8 in the form

d2 R 1 dR 2µ W 2 V0
(9) + + 2 2 + − V0 R .
dy 2 y dy α ~ y2 y
R(y) must be finite, continuous and single valued in the range 0 ≤ y ≤ ∞. Equation (9) can be
solved if the following guess for the y dependence is made R(y) = e−dy (2dy)b/2 F (y). We obtain
dR b
(10) = e−dy (2dy) 2 −1 {F (y)[bd − d(2dy)] + (2dy)F 0 (y)} ,
dy
and
d2 R b
(11) 2
= e−dy (2dy) 2 −2 ·
dy
2 00
{(2dy) F (y) + (2dy)[2db − 2d(2dy)] F 0 (y) + [d2 (2dy)2 − 2d2 b(2dy) + d2 b(b − 2)]F (y)} .
After substitution in Eq. (9), we obtain
b 2µV0

(12) e−dy (2dy) 2 −2 (2dy)2 F 00 (y) + (2dy)2d[b + 1 − 2dy]F 0 (y) + (2dy)2 d2 − α 2 ~2
d2

−2d(2dy) db + d − 4µV 0
α 2 ~2
+ d2 b2 + 8µW
α 2 ~2
F = 0,
For
α 2 b 2 ~2
(13) W =− ,
8µ
2
the two terms d2 b2 and 8µW d
α 2 ~2
in Eq. (12) cancel each other. A further coordinate transformation,
namely z = 2dy, leads to
(14)    
 2

z b d F dF 2µV0 2µV0 b 1 
e− 2 (z) 2 −1 z 2 (z) + [b + 1 − z] (z) + z d2 − 2 2 + − − F (z) = 0.
 dz dz α ~ α 2 ~2 d 2 2 
| {z }
2.1. THE SCHÖDINGER EQUATION FOR THE MORSE POTENTIAL 13
1
(2µV0 ) 2
The under-braced term vanishes if we choose d = α~
, as we are free to do. Eventually the
following equation for F (z) is obtained:

d2 F dF b 1
(15) z 2 (z) + [b + 1 − z] (z) + d − − F (z) = 0 .
dz dz 2 2

Equation (15) allows a polynomial as solution if d − 2b − 21 = n, n integer > 0, i.e. if
1 1
2(2µV0 ) 2 2(2µV0 ) 2
(16) b= − 1 − 2n = k − 1 − 2n, k ≡ ≡ 2d .
α~ α~
A discrete spectrum is obtained provided that k > 1.
Taking into account all the coordinate transformations employed so far, the complete general
solution for the radial part of the 3D Schrödinger equation for the Morse potential can be written
explicitly as
k −α(r−r0 ) k−1−2n
(17) Rn (r) = Nn e− 2 e (ke−α(r−r0 ) ) 2 Lk−1−2n
n (ke−α(r−r0 ) ) ,
where Lνn (x) are the generalized Laguerre Polynomials, which can be represented in differential
form as
ex x−ν dn ν+n −x
(18) Lνn (x) = (x e ) .
n! dxn
Recalling the structure of the complete eigenfunction (5), as the angular part is assumed
normalized, the normalization condition in 3D is written as
Z ∞
(19) 0
hΨ|Ψ i = r 2 Rn (r)Rn∗ 0 (r)dr = 1 ,
0
explicitly
Z αr0
N 2 ke 0 0
(20) 0
hΨ|Ψ i = − dze−z z k−n−n −2 Lk−1−2n
n (z)Lk−1−2n
n0 (z) .
α 0
Useful relations of standard handbooks (e.g. Ref. [25]) allow the explicit evaluation of several
integrals involving Laguerre polynomials (see Appendix 9.1). We use in particular:
Z ∞
µ0
X λ − µλ − µ0 λ + k
λ −z µ n+n0
(21) z e Ln (z)Ln0 (z)dz = (−) Γ(λ + 1) 0 −k
.
0 k
n − k n k
Equation (21) will be largely employed in the following to evaluate scalar products, and the
matrix elements of most operators. Notice that the upper integration limit for the z variable in
Eq (20) is supposed to be keαr0 , due to the fact that the radial variable r belongs to [0, +∞).
The use of formulas such as (21), where integration extends to z → +∞, i.e. to r → −∞
introduces a small error in the evaluation of the integrals based on Eq. (21). Nonetheless for
practical values of parameters, this error is extremely small, and can be neglected safely. This
same inconvenient occurs in the case of a hypothetical 2D Morse problem, while the 1D Morse
problem does not suffer from this error, as there the one dimensional spatial variable is defined
on the whole real axis (see Appendix 9.2).
For future reference, we introduce a more compact notation, that will be largely employed
in the following chapters. We write the Schrödinger equation for the Morse potential (4) as
2
p̂x
(22) ĤM Ψ(x) ≡ + VM (x) Ψ(x) = W Ψ(x) ,
2µ
and the Morse Potential function can be written as

(23) VM (x) ≡ V0 [v(x)]2 − 1 , with v(x) = e−α(x−x0 ) − 1 .
The three quantities parameterizing the potential function are V0 (depth of the well), α (shape
parameter), and x0 (position of the minimum). For later convenience we introduce the dimen-
sionless combination s
√
2mV0 1 1
(24) s= − ≡d− ,
~α 2 2
also related to k(= 2d) by the relation
(25) 2s + 1 = k .
The bound-state eigenfunctions can be written as
(26) Ψn (x) = Nn e−y(x)/2 y(x)s−n Ln2(s−n) (y(x)) ,
where
(27) y(x) = (2s + 1) e−α(x−x0 ) .
Lνn (z) are the Laguerre polynomials defined in Eq. (18). The normalization constant evaluated
through Eq. (21) is
1
1/2 Γ(2s − n + 1) 2
(28) Nn = α .
2(s − n)Γ(n + 1)
For integer n = 0, 1, 2, . . . [s] ([s] indicates the largest integer ≤ s), the square-integrable wave-
functions Ψn (x) represent the [s] + 1 bound states, for the specific physical parameters of the
system. The corresponding energy eigenvalues are
α 2 ~2
(29) Wn = − (s − n)2 .
2µ
2.2. Molecular potential approximation
The discussion in the previous section shows that Morse potential yields an exactly-solvable
Schrödinger problem, and possesses the overall shape of a realistic molecular vibrational poten-
tial. Beside its many advantages, the Morse potential has a few drawbacks: mainly, it does not
approach the dissociation limit (here E = 0) with a power law as realistically expected, and,
less importantly, it does not quite diverge at x = 0.
Nowadays’ experimental spectroscopy allows accurate determination of energy levels at the
higher end of the bound-state energy range, very close to dissociation. A theoretical approach
which aims at being really applicable in that range must then face many problems. In par-
ticular, a Hamiltonian description of the molecular problem needs a very accurate potential,
because minor inaccuracies in the potential function produce great differences in the computed
spectrum, mainly close in energy to dissociation. A more realistically accurate potential V (x)
can in principle be obtained by adding a suitable correction Vd (x) ≡ V (x)−V0 (x) to an arbitrary
starting potential V0 (x). Two main considerations can be made.
First, to develop a method to solve the problem based on the physical properties rather
than on brute force, it is advantageous to choose a starting potential V0 (x) whose Schrödinger
problem is exactly solvable, in order to obtain analytic expressions of basis wavefunctions and
eigenvalues.
Second, the correction Vd (x) = V (x) − V0 (x) should be in some sense small with respect to
V0 (x), in the sense that the Hilbert space decomposition related to V0 (x) retains most of its
significance for the complete Hamiltonian problem. The correction Vd (x) is often written as a
suitable expansion, whose detailed form is determined by the choice of V0 (x).
The determination of the best functional form of the starting approximate molecular poten-
tial is not trivial: it involves a compromise between the accuracy of the potential approximation
2.2. MOLECULAR POTENTIAL APPROXIMATION 15
and the manageability of the resulting set of equations. An exact functional representation of
the potential is of no use if the corresponding equation is manageable only through a heavy
numerical load.
In the past many approaches have been explored, imposing the model potential in the form
of a (power) series in suitable variables. One well known expansion is the power series in the
displacement from equilibrium coordinate:
N
X max
(30) V (x) = ai (x − x0 )i ,
i=2
where V0 (x) = a2 (x − x0 )2 is taken as the starting potential. The assumed underlying basis is
that of the harmonic-oscillator eigenvectors. The correction term is
N
X max
(31) Vd (x) = ai (x − x0 )i .
i=3
This traditional expansion is usually treated by perturbation theory. However, for large x the
expansion (31), and thus Vd (x), diverges to +∞ or even to −∞ , and so cannot describe well the
whole energy range of a realistic molecular potential. In particular, such expansion is guaranteed
to provide poor results where the real potential approaches dissociation, as illustrated in the
Introduction.
Other kinds of expansion have been proposed and used. For example, the variable y = x−x x0
0
(see for example Ref. [4]) shows better convergence properties. Alternatively, the Simons-Parr-
Finlan coordinate y = x−x x
0
[5], or a similar coordinate y = 2(x−xx+x0
0)
introduced in Ref. [15],
provide a physically reasonable expansion for both small and large x.
Taking advantage of the stronger similarity of the Morse potential to a realistic potential,
an expansion in terms of Morse coordinate has been suggested and employed, in diatomic and
polyatomic context (see Chapter 6). The potential expansion
X
(32) V (r) = VM (x) + bi [v(x)]i ,
i=4
which shows better convergence properties than Eq. (31), is the core of the Perturbed Morse
Oscillator (PMO) method [26, 27, 28, 29, 30].
The exact form of the expansion is chosen by taking care of three main requirements:
(a) the expansion must be sufficiently flexible to allow an accurate approximation of a
generic potential,
(b) every term in the expansion must have easily and accurately computable matrix ele-
ments in the chosen Hilbert-space basis, and
(c) the matrix representation has better be sparse for more efficient diagonalization (man-
ageability).
Once these three basic requirements are satisfied, a complete Hamiltonian study of the corre-
sponding physical problem is carried out in three steps:
• determine the weights in the expansion of the potential so that it best approximates
the true V (x);
• construct the matrix representation of V (x);
• diagonalize it to obtain eigenvalues and eigenfunctions.
The harmonic-oscillator potential does not satisfy requirement (a) above. It can approxi-
mate V (x) arbitrarily well in a neighborhood of x0 , but no number of polynomial terms can
provide a meaningful approximation of V (x) as it approaches dissociation. As the matrix ele-

ments are very simple analytic expressions, the harmonic-oscillator based expansion is extremely
convenient in terms of manageability, but lacks in flexibility.
2.3. Expansion in the Morse basis

The method we propose here is based on an expansion for V (x) very much like Eq. (32).
The difference is the introduction of a [v(x)]3 term. We take therefore
N
X max
(33) Vd (x) = ai [v(x)]i ,

i=3
which for the complete potential implies the expanded form

N
X max
(34) V (x) = ai [v(x)]i , where a2 ≡ V0 .

i=2
Initially, we expand the general eigensolution of the Schrödinger equation for the potential
V (x) = VM (x)+Vd (x) on the set of the bound eigenstates of the underlying Morse Hamiltonian.
In this basis, the correction potential Vd (x) of Eq. (33) allows exact evaluation of its matrix
elements. It therefore satisfies the criterium of formal manageability. Chapter 3 below deals
with the problem of flexibility, but we anticipate that the form (33) is indeed very flexible, and
locally accurate, thanks to the introduction of the [v(x)]3 term.
As Eq. (21) suggests, (see also Appendix 9.1) matrix elements of [v(x)]i involve the Laguerre
polynomial integral relation. It is relatively easy to find that matrix elements of any power of
[e−α(x−x0 ) ]t (t > 0) in the basis (26) can be evaluated exactly and analytically. Matrix elements
of potential terms on Morse bound states, either exact or approximate, are published in a
number of papers in literature [6, 31, 32]. We report here an explicit expression for the terms
that we are interested in.
In the basis of Morse bound-state wavefunctions of the form (26)
Ψn (x) = Nn e−y(x)/2 y(x)s−n Ln2(s−n) (y(x)) , y(x) = (2s + 1) e−α(x−x0 ) ,
the general Laguerre polynomials integral formula (21) allows to write the matrix element
hn|[e−a(r−r0 ) ]t |mi between Morse eigenstates (26) (see Appendix 9.1) as
Z ∞
−a(r−r0 ) t Nn Nm
(35) hn|[e ] |mi = − t
dye−y y 2s−n−m−1+t L2s−2n
n (y)L2s−2m
m (y) ,
α(2s + 1) 0
where we neglect the small correction in normalization due to the boundary conditions in 2-3D.
We obtain
(−1)n+m
hn|[e−α(x̂−x0 ) ]t |mi = Nn Nm Γ[2s − m − n + t] ×
α(2s + 1)t
min(n,m)
X n − m − 1 + t m − n − 1 + t 2s − 1 − m − n + ` + t
(36) ,
`=0
n−` m−` `
where Nn , Nm are the normalization constants of Eq. (28). Equation (36) holds for real t ≥ 0,
but we are interested only in integer t: in this case, the first non vanishing term in the sum
occurs for ` = (1 − δn,m ) max(0, min(n, m) + 1 − t). The matrix elements of any power term
[v(x)]i can be easily computed analytically by combining expressions of the kind provided by
Eq. (36). Note that the matrix element (36) is non vanishing for any n, m: therefore the
2.3. EXPANSION IN THE MORSE BASIS 17
Figure 2.2. Typical matrix elements behavior of hn|[v(x)]4 |mi in the basis of
Morse bound states. The grey intensity is proportional to the absolute value of
the matrix elements, and the position along the sides of the square represents
precisely the indexes n and m of the involved Morse states.
resulting matrix is not sparse, as shown pictorially in Fig. 2.2. The resulting representation
therefore violates the requirement (c) of Sec. 2.2.
Equation (36) permits the construction of the matrix representation of the Hamiltonian
(37) H = HM + Vd (x) ,
which is exact within the space of Morse bound states. The molecular problem is thus reduced
to the diagonalization of a ([s] + 1) × ([s] + 1) matrix: clearly it yields exact eigenvalues and
eigenvectors in the trivial limit H → HM (i.e. Vd (x) = 0).
When Vd (x) 6= 0, H mixes states within the (finite) Morse basis to states orthogonal to
it. The matrix representation of H is therefore only approximate, and provides a variational
estimate of the energy of the lowest bound states. Due to the imperfect overlap of the subspace
of Morse bound states to the subspace of exact bound states of H, we expect that when the
correction Vd (x) to the Morse problem becomes important in Eq. (37), the accuracy of the
description deteriorates.
2.3.1. The problem of basis incompleteness. We have not yet proposed a general
recipe to obtain the best approximation of the form of Eq. (33) to a given microscopic potential:
Chapter 3 is devoted to a detailed discussion on this crucial problem. For now, we assume here
that an accurate approximation of a molecular 1D potential through an expansion (33) is
available. We concentrate here on the problem of the finiteness of Morse potential bound states
basis: this attractive feature is in fact a great obstacle to the application of this basis to an
arbitrary potential.
As a completely general argument we may say that if {ϕVn0 }n=0,1,...,N̂ is a finite-dimensional
subspace of solutions of a H0 Hamiltonian, an additional potential term Vd will in general lead
to solutions that are a mixture of those of the original problem. In the most favorable cases
the effect is only to mix the original Hamiltonian eigenstates, but in general the solutions of
the new Hamiltonian will have non-vanishing components on the Hilbert subspace orthogonal
to {ϕVn0 }n=0,1,...,N̂ , that of unbound states for the Morse oscillator. In such a situation, the
R = 0.01
R = 0.02
-0.5 R = 0.20
R = 1.0
Morse
Energy [V0]
-1
-1.5
-2
1 2 4 6 8 10
αx
Figure 2.3. Pure Morse potential and potential V (x) = VM (x) + a4 [v(x)]4 −
(V0 + a4 ), for a4 = `V0 , ` = 0.01, 0.02, 0.20 and 1.0. The potentials are shifted in
order to have the dissociation limit equal to 0. (~ = 1, µ = 1, s = 8.34).
accuracy of the description of the system H through the basis of H0 is clearly questionable,
and indeed this approach can be considered reliable only for relatively small correction V d .
This is an intrinsic problem affecting any basis related to potentials whose Schrödinger
equation leads to a finite number of bound states, and is not of easy solution, as the simultaneous
treatment of the bound and continuum part of the spectrum makes the whole treatment more
intricate [33, 34].
This problem appears also in more favorable situations when a realistic computation is
involved. In the best known and most commonly employed approach, the harmonic-oscillator-
based expansion, in principle a complete basis of eigensolutions is available, but in practical
calculations one must make some basis truncation and consider a finite-dimensional subspace,
and so (artificially) decompose of the whole Hilbert space in two orthogonal subspaces. It is
the same incompleteness problem of the Morse basis with the important difference that the
accuracy of the description here is not determined by the starting Hamiltonian, and can be
improved systematically.
As a concrete case consider Vd (x) = a4 [v(x)]4 , so that the complete potential is
(38) V (x) = VM (x) + a4 [v(x)]4 − (V0 + a4 ) .
We choose parameters V0 = 625, α = 4, x0 = 1, µ = 1 and ~ = 1, so that s = 8.34, and thus
the pure Morse potential allows [s] + 1 = 9 bound states.
Define the dimensionless parameter R, that gauges the overall weight of correction terms
Vd , as follows:
Nmax
1 X
(39) R= |ai | .
V0 i=3
In the case at hand R = a4 /V0 . We solve the Hamiltonian problem diagonalizing the corre-
sponding 9×9 matrix, constructed using the relations (36), for different values of the parameter
a4 = `V0 , namely ` = 0.01, 0.02, 0.2. In this particularly simple example, the parameter ` ≡ R
itself measures the depth of the correction term with respect to the Morse potential.
state Finite Differences Morse basis

0 -562.482 -562.482
1 -436.839 -436.839
2 -326.998 -326.998
3 -232.987 -232.985
4 -154.832 -154.829
5 -92.560 -92.557
6 -46.198 -46.187
7 -15.772 -15.729
8 -1.310 -1.253
Table 2.1. Eigenvalues computed by diagonalizing the Schrödinger Hamiltonian
corresponding to the potential of Eq. (38) for R = Va40 = 0.01, compared to exact
eigenvalues obtained by finite differences. The total discrepancy is ∆9 = 0.009%.

0 -568.674 -568.674
1 -442.812 -442.811
2 -332.555 -332.553
3 -237.960 -237.954
4 -159.080 -159.070
5 -95.968 -95.957
6 -48.678 -48.632
7 -17.260 -17.084
8 -1.766 -1.550
Table 2.2. Same as Table 2.1, for R = 0.02. The total discrepancy is ∆9 = 0.03%.
Tables 2.1 and 2.2 report the eigenvalues obtained. As no analytical exact solution for the
complete problem is available, we use for a comparison the results obtained through a fully
converged numerical computation based on a finite-difference algorithm. Note that in all cases
the dissociation threshold is set to zero, in order to identify easily the bound states. For all the
chosen values of the parameter a4 , the complete potential well is deeper than the corresponding
Morse one, as seen in Fig. 2.3.
To make our considerations quantitative, we introduce the estimator
1
(40) ∆Ns := max |Ekalg − Ekex | ,
V0 k
where Ekalg are the eigenvalues obtained diagonalizing the matrix, and Ekex are the exact eigen-
values computed with the finite differences method. Ns = 9 indicates the number of states
of the basis considered: for the moment it is a redundant quantity, but it will be used in
subsequent sections.
The value of ∆9 changes from ∆9 = 0.009% for R = 0.01 to ∆9 = 2.5% for R = 0.20.
For small relative weight the accord is rather good, but it reduces rapidly. Already for the
correction term at 20% of the Morse one (Table 2.3), the situation changes significantly: the
complete potential is sufficiently deep and wide to allow an additional bound state.
Notice incidentally that according to the previous analysis the lack of accuracy due to the
finiteness of Morse basis affects almost exclusively the states close to dissociation, while the
lowest lying states are still described very accurately. This consideration remain valid even for

0 -680.179 -680.169
1 -550.549 -550.494
2 -433.313 -433.121
3 -329.098 -328.533
4 -238.424 -237.481
5 -161.742 -159.926
6 -99.456 -92.210
7 -51.940 -37.126
8 -19.540 -4.866
9 -2.604 (missing)
Table 2.3. Same as Table 2.1, for R = 0.20. The total discrepancy is ∆9 =
2.5%, and one bound state is missing.
1
R=0.02
0.9 R=0.20
Projection
R=0.50
0.8
0.7
0.6
0 5 10
eigenstate k
P
Figure 2.4. The projection [s] 2
j=0 |hΨj |ϕk i| of the k
th
eigenstate on the sub-
space spanned by the Morse basis, for increasing relative importance R of the
deformation Vd (x) added to a Morse potential characterized by [s] + 1 = 9 bound
states (~ = 1, µ = 1, s = 8.34).
correction terms corresponding to more complex expansions and for larger relative weight. We
will come back to this point in greater detail in Chap. 5. The full Hamiltonian H mixes states
within the Morse basis to states belonging to the orthogonal space, and clearly the method itself
loses accuracy. This can be made evident by analyzing the components of the eigenfunction of
the complete potential in Morse basis.
We compute the squared projection of the k th eigenstate obtained via finite differences
method ϕk on Morse basis (26):
[s]
X
(41) Projection = |hΨj |ϕk i|2 .
j=0
Table 2.4 and Figure 2.4 illustrate the results. Increasing the weight of correction term the
mixing to states orthogonal to Morse basis increases, and the total projection becomes smaller
than unity. This affects mainly high-energy states, and evolves monotonically with the increase
of the correction weight, at least until additional states appear. The evidence coming from
eigenvalues analysis is thus confirmed.
k R = 0.01 R = 0.02 R = 0.20 R = 0.50

0 1. 1. 0.999994 0.999974
1 1. 0.999999 0.999962 0.999844
2 0.999999 0.999998 0.999833 0.999134
3 0.999998 0.999993 0.999338 0.996245
4 0.999996 0.999985 0.998762 0.993674
5 0.999995 0.999978 0.995127 0.95088
6 0.999921 0.99966 0.946333 0.761578
7 0.999116 0.996468 0.799495 0.705957
8 0.992598 0.974095 0.750018 0.714221
9 ———— ———— 0.891871 0.722432
10 ———— ———— ———– 0.702376
11 ———— ———— ———– 0.668112
P
Table 2.4. The projection [s] 2
j=0 |hΨj |ϕk i| of the eigenstates on the subspace
spanned by the Morse basis, for increasing relative weight R.
1 8
7
0.5
6 R=1
5
Energy [V0]
0 4
3
13
2 8 12
2
-0.5 11
1 1
0 0 10
-1 7
2 4 6 8 10
αx
Figure 2.5. Morse potential (dashed) and potential expansion with Vd (x) =
1.0 V0 [v(x)]4 (solid). Solid horizontal lines represent the exact (finite differences)
eigenvalues of this potential well, while dashed horizontal lines indicate the ap-
proximate eigenvalues evaluated within the Morse basis.
For future reference we show here potential and eigenstates for two simple highly distorted
potentials: Vd (x) = V0 [v(x)]4 (Fig. 2.5) and Vd (x) = V0 [v(x)]6 (Fig. 2.6), i.e. both corre-
sponding to R = 1. The effect of the finiteness of the Morse basis is substantial: the corrected
potential allows 14 and 15 bound states respectively, while pure Morse potential provides only a
9-dimensional subspace. Here we follow a different convention, and the zero of the energy scale
is chosen to be the Morse-term dissociation energy. The corresponding eigenvalues obtained by
diagonalizing the Hamiltonian in the Morse basis are reported in Tables 2.5 and 2.6. As can
be seen from the spectra, in both cases when the correction weight is so large, Morse basis fails
completely: several states are completely missing, and only few low-lying states are found with
an accuracy better than 2% of V0 .
1 8
7
0.5 6 R=1
Energy [V0]
5
0 4
3
8 13 14
-0.5 2 2 12
1 1
11
0 0
-1
2 4 6 8 10
αx
Figure 2.6. Morse potential (dashed) and potential expansion with Vd (x) =
1.0 V0 [v(x)]6 (solid). Solid horizontal lines represent the exact (finite differences)
eigenvalues of this potential well, while dashed horizontal lines indicate the ap-
proximate eigenvalues evaluated within the Morse basis.

0 -551.35 -551.23 (0.019%)
1 -407.92 -407.27 (0.10%)
2 -266.82 -263.00 (0.61%)
3 -131.4 -117.73 (2.2%)
4 -3.76 19.21 (3.7%)
5 114.66 178.64 (10%)
6 222.75 361.3 (22%)
7 319.5 517.6 (32%)
8 404.1 609.7 (33%)
9 476.0
10 534.5
11 578.9
12 608.8
13 623.5
Table 2.5. Potential expansion with Vd (x) = 1.0 V0 [v(x)]4 , (α = 4, x0 =
1, V0 = 625, µ = 1, ~ = 1): comparison of results of Morse basis (Ns = 9)
and finite differences method. In brackets the differences |Ekalg − Ekex | in percent
of V0 .
Table 2.7 shows the eigenvalues computed for the correction potential (Fig. 2.7)
6
X
(42) Vd (x) = 2.0 V0 [v(x)]i .
i=3
This correction corresponds to an overall factor R = 8, Eq. (39), but the well depth is the
same as pure Morse. The overall shape is really different from Morse. Nonetheless, this is a
well-behaved potential through the whole range in the x variable, and one would like to be able
to treat it accurately. However, the Morse basis of course performs particularly poorly, in this

0 -554.87 -554.74 (0.022%)
1 -422.79 -422.10 (0.11%)
2 -296.42 -294.06 (0.38%)
3 -174.25 -164.81 (1.5%)
4 -56.87 -36.73 (3.2%)
5 54.67 104.84 (8%)
6 159.20 295.77 (22%)
7 255.64 484.1 (37%)
8 342.98 604.4 (42%)
9 420.2
10 486.5
11 540.9
12 582.6
13 610.6
14 623.9
1, V0 = 625, µ = 1, ~ = 1): comparison of diagonalization within the Morse ba-
sis (Ns = 9) and finite differences method. In brackets the differences |Ekalg −Ekex |
in percent of V0 .
0
Energy [V0]
-0.5
-1
1 5 9
αx
Figure 2.7. The Morse potential (dashed) compared to the potential with a
P
correction term Vd (x) = 2.0 V0 6i=3 [v(x)]i (solid).
case: the accuracy is low even for low-lying states, and it even produces 8 bound states only
instead of 11.
These examples could suggest that expansion in Morse basis works worse than one could
expect. These examples were purposely chosen to show the effect of the limitations of the
method in a general situation. Eventually, the Morse basis performs well in at least two
situations:
• for small correction potential Vd ,
• when Morse-term basis size is greater than the number of bound states of the distorted
potential (see Chap. 4 below).

0 -567.69 -567.53 (0.025%)
1 -493.33 -493.01 (0.052%)
2 -415.47 -414.14 (0.21%)
3 -337.81 -336.10 (0.27%)
4 -263.62 -258.23 (0.86%)
5 -195.33 -186.24 (1.45%)
6 -134.84 -92.43 (6.8%)
7 -83.77 -14.35 (11%)
8 -43.58 (652.77) (111%)
9 -15.68
10 -1.50
P
Table 2.7. Potential expansion with Vd (x) = 2.0 V0 6i=3 [v(x)]i , (α = 4, x0 =
1, V0 = 625, µ = 1, ~ = 1): comparison of diagonalization within the Morse basis
(Ns = 9) and finite differences method. In brackets the differences |Ekalg − Ekex |
in percent of V0 .
As the examples show, though, the Morse-basis finiteness can prevent an accurate treatment
of the general oscillator.
2.4. Beyond the Morse basis

The previous examples show that the finiteness of Morse bound states basis is an obstacle
to the construction of a really general method to treat molecular vibrations. To obtain high
accuracy throughout the whole energy range, particularly in the tough region close to dissocia-
tion, it is necessary to use a basis which can be systematically enlarged, virtually to cover the
entire Hilbert space, like the eigenfunctions of the harmonic oscillator.
Of course, given any Hamiltonian, one could in principle diagonalize it in an arbitrary
complete orthonormal system in the appropriate Hilbert space, but this is usually a bad choice
for a numerical diagonalization, as it violates requirements (b) and (c) of Sec. 2.2. Obviously
an analytical solution would take great advantage from a previous knowledge that could lead
to an a priori choice of a basis to make the Hamiltonian matrix already diagonal, or at least
particularly simple. This is possible, but only in a limited number of textbook problems.
The situation is particularly complicated when the spectrum of the Hamiltonian operator
involved is made of a finite-dimensional discrete spectrum plus a continuum spectrum. The
inclusion of the continuum is a complicated problem by itself, but a few reasonable approximate
methods have been developed.
The quest for manageable approximations of the continuum spectrum has followed different
routes, including analytical methods (wavefunctions manipulation in coordinate representation
[33, 34]), to essentially algebraic treatments related to the dynamical-symmetry viewpoint [35]
or of supersymmetric (SUSY) quantum mechanics derivation [36].
In the present work we follow the route of completing the set of Morse eigenstates by
replacing it with the quasi number state basis (QNSB), which is related to supersymmetric
(SUSY) quantum mechanics.
2.5. Morse and SUSY Shape Invariant Potentials

2.5.1. The QNSB and consequences. To replace the incomplete Morse bound states
set, a different but related set of functions is chosen with great care to reduce computational
2.5. MORSE AND SUSY SHAPE INVARIANT POTENTIALS 25
efforts. We choose to follow an algebraic approach as it is the most suited for a subsequent
generalization to potentials of the kind of Eq. (33). In addition, if an algebraic form of the
dynamical variables is available, the computation of matrix elements and other interesting
quantities can be done by exploiting simple algebraic relations. We sketch here the main ideas
followed in the literature, with particular attention to what is useful for our specific work.
The basic idea is to build appropriate generalized ladder operators connecting the different
wavefunctions of a given family, and subsequently express all relevant quantities in terms of
these operators. The ladder operators, on one side, allow the identification of the dynamical
symmetry of the physical model under consideration, and, on the other side, lead to an al-
gebraic expression for all relevant dynamical variables, including the Hamiltonian. A similar
construction was made also for the Morse potential [37], starting from the Morse bound-state
eigenfunctions (26).
Eventually, we need a basis where suitable ladder operators act: this basis must be complete
in L2 [(0, ∞), dy/y]. To this purpose we introduce the following family of wavefunctions [35]
s
αn! y
(43) φσn (y) = y σ e− 2 Ln2σ−1 (y), n = 0, 1, 2, . . . ,
Γ(2σ + n)
characterized by a real parameter σ > 0. Following Ref. [14], we refer to |φσn i as a quasi
number state. Formally, these functions resemble Morse bound states (26). However, they are
substantially different. In particular for any σ > 0, the quasi number states do form a complete
orthonormal set in L2 [(0, ∞), dy/y] [35]. This same set was introduced in Ref. [14], on the basis
of considerations related to SUSY quantum mechanics, and applied to the context of molecular
spectra in Ref. [9, 11].
The construction of generalized ladder operators realized for Morse bound states can be
repeated starting from Eq. (43) [35]. The three resulting operators are

ŷ d ŷ i
K̂− = (σ Iˆ + n̂) − − y = (σ Iˆ + n̂) − + p̂x
2 dy 2 ~α

ˆ ŷ d ˆ ŷ i
(44) K̂+ = (σ I + n̂) − + y = (σ I + n̂) − − p̂x
2 dy 2 ~α
K̂0 = σ Iˆ + n̂ ,
where n̂φσn = nφσn . K̂− , K̂+ and K̂0 act as generalized step up, step down and number operator
repeatedly on the states represented by Eq. (43):
(45) K̂− |φσn i = Cn |φσn−1 i
(46) K̂+ |φσn i = Cn+1 |φσn+1 i
(47) K̂0 |φσn i = (σ + n)|φσn i ,
where
p
(48) Cn = n(n + 2σ − 1) .
The definition (44) can be inverted, to obtain the operator form of momentum and Morse
coordinate
i~α
(49) p̂x = (K̂+ − K̂− )
2
(50) ŷ = (2s + 1)e−α(x̂−x0 ) = 2K̂0 − (K̂+ + K̂− ) .
Accordingly, the whole Hamiltonian (22) can be expressed in algebraic form, as done in Ref. [35]:
the matrix elements hφσm |ĤM |φσn i can be computed algebraically. The only non-vanishing matrix
elements are those for which m = n, n ± 1: the Morse Hamiltonian matrix is tridiagonal in the
basis |φσn i. All the above properties hold regardless of the value of σ, for which a convenient
choice will be made in Sect. 2.5.5.
2.5.2. SUSY. Super Symmetric quantum mechanics is a minor daughter of the large scale
attempt to reach unification of physical fundamental forces. The intricacies of this kind of
theory go beyond the scope of this work, and somewhat beyond the critical capacity of the
author. We only note that a lot of sophisticated field theory methods have been devised, whose
correctness often is not easy to assess.
SUSY quantum mechanics was originally born as a testing ground for verifying the validity
of the general methods in a controlled environment in which (some) exact results were available.
After the initial attempts it was realized that SUSY QM is an interesting field in its own, and
in last 15 years several important results have been obtained. It is not our purpose to be
exhaustive, but we think that the general SUSY QM approach is very useful to the problem at
hand, and we devote some space to sketch its main ideas. In doing so, we follow the interesting
and accurate review of Ref. [36].
SUSY QM Hamiltonian formulation starts observing that in 1D there is a rigorous connec-
tion between the bound-state wavefunctions and the potential: once one knows the ground-state
wavefunction (or any other bound state), one knows exactly the potential (up to a constant).
Assume ϕ0 (x) is the ground-state wavefunction of some Hamiltonian Ĥ1 whose eigenenergy
equals 0:
~2 d2 ϕ0 (x)
(51) Ĥ1 ϕ0 (x) = − + V̂1 (x)ϕ0 (x) = 0 ,
2m dx2
so that
~2 ϕ000 (x)
(52) V̂1 (x) = .
2m ϕ0 (x)
This equality allows the reconstruction of the potential V̂1 (x) from its ground-state wavefunc-
tion, as long as it has no nodes. Once this is realized, it is possible to try and factorize the
Hamiltonian through the ansatz
(53) Ĥ1 = Â† Â ,
where
~ d ~ d
(54) Â = 1/2
+ W (x), Â† = − 1/2
+ W (x) .
(2m) dx (2m) dx
This leads to an equation for W (x), that turns out to be a Riccati equation:
~
(55) V̂1 (x) = W 2 (x) − W 0 (x) .
(2m)1/2
In SUSY language W (x) is usually referred to as superpotential. From Eq. (52), W (x) can be
expressed in terms of the ground-state wavefunction:
~ ϕ00 (x)
(56) W (x) = − ,
(2m)1/2 ϕ0 (x)
as can be seen recognizing that if Âϕ0 (x) = 0 then Ĥ1 ϕ0 (x) = Â† Âϕ0 (x) = 0.
SUSY recipe is now to construct the Hamiltonian
(57) Ĥ2 = ÂÂ† ,
Figure 2.8. Energy levels of two supersymmetric partner potentials V̂1 and V̂2
(unbroken SUSY). Energy levels are equal, except that V̂1 has an extra state at 0
energy. Operators Â and Â† connect eigenstates belonging to different potentials.
obtained reversing the order of Â operators in Ĥ1 , that may be seen explicitly as related to a
new potential V̂2 (x):
~2 d 2 ~
(58) Ĥ2 = − + V̂2 (x), V̂2 (x) = W 2 (x) + W 0 (x) .
2m dx2 (2m)1/2
The two potentials V̂1 (x) and V̂2 (x) are known as supersymmetric partner potentials. The
two corresponding Hamiltonians Ĥ1 and Ĥ2 show a strict connection, and the eigenvalues and
eigenfunctions are strictly related. For n > 0, the Schrödinger equation for Ĥ1
(59) Ĥ1 ϕ(1) † (1) (1) (1)
n = Â Âϕn = En ϕn
implies
(60) Ĥ2 (Âϕ(1) † (1) (1) (1)
n ) = ÂÂ (Âϕn ) = En (Âϕn ) ,
and reversely
(61) Ĥ2 ϕ(2) † (2) (2) (2)
n = ÂÂ ϕn = En ϕn
implies
(62) Ĥ1 (Â† ϕ(2) † † (2) (2) † (1)
n ) = Â Â(Â ϕn ) = En (Â ϕn ) .
So the eigenvalues and eigenfunctions of the two partner Hamiltonians satisfy the relations:
(1) (1)
(63) En(2) = En+1 , E0 = 0 ,
(1) (1)
(64) ϕ(2)
n = [En+1 ]−1/2 Âϕn+1 ,
(1)
(65) ϕn+1 = [En(2) ]−1/2 Â† ϕ(2)
n .
The operator Â (Â† ) converts an eigenfunction of Ĥ1 (Ĥ2 ) into an eigenfunction of Ĥ2 (Ĥ1 )
of the same energy, and destroys (creates) one node in the eigenfunction. In this scheme, the
ground state of Ĥ1 has no SUSY partner, as it is annihilated by Â. When this is the case, it is
said that SUSY is unbroken, and we will consider only such a case.
In summary, if all the eigenfunctions of Ĥ1 are known, then the eigenfunctions of Ĥ2 can
be obtained through the operator Â, and conversely Â† allows to obtain all the eigenfunctions
of Ĥ1 Hamiltonian, except for the ground state, if all eigenfunctions of Ĥ2 are known (Figure
2.8).
PSfrag replacements
Â1 Â2 Ân−1
Ĥ1 Ĥ2 ··· Ĥn

Figure 2.9. Hierarchy of Hamiltonians.
Another crucial though simple concept is that of hierarchy of Hamiltonians. Given the
ground state of the Hamiltonian Ĥ1 , through Eq. (56) the corresponding superpotential W1 (x)
is completely specified, and the operators Â1 and Â†1 give the factorized Ĥ1 via Eq. (53). But the
ground state of the partner Hamiltonian Ĥ2 is also known through the ground state of Ĥ1 and
the operator Â1 . The Ĥ2 Hamiltonian can in turn be factorized by means of a superpotential
W2 . This produces a partner Hamiltonian Ĥ3 . This procedure can be repeated to obtain what
is named in SUSY a hierarchy of Hamiltonians (Figure 2.9). Each member has one bound state
less than its predecessor, so the process stops when the number of bound states is exhausted.
For example, if the Ĥ1 Hamiltonian corresponds to an exactly solvable problem, then the
eigenvalues and eigenfunctions of all the Hamiltonians of the hierarchy can be obtained through
successive factorization. If instead the ground states of all the Hamiltonians are known, the
solutions of the general problem are automatically available.
The concept of SUSY hierarchy is particularly useful in combination with another peculiar
SUSY concept, the Shape Invariance [38, 39]. If the two partner potentials V̂1,2 are similar in
shape and differ only in the parameters that are present in their definition, they are said to be
shape invariant (Shape Invariant Potential, also referred as SIP in the following). In formulas,
they must satisfy the condition
(66) V̂2 (x; a1 ) = V̂1 (x; a2 ) + R(a1 ) ,
where a1 is a set of parameters, a2 = f (a1 ), and R(a1 ) does not depend on x.

The Shape Invariance condition is an integrability condition: if Eq. (66) is satisfied, for
unbroken SUSY the energy eigenvalues and eigenfunctions of any SIP can be obtained alge-
braically. In detail, if Ĥ1 and Ĥ2 are two SUSY partner hamiltonians, we can write
Z x
(1) (1)
(67) E0 (a1 ) = 0, ϕ0 (x; a1 ) = N exp − W1 (y; , a1 )dy .
Starting from Ĥ1 construct the hierarchy of Hamiltonians Ĥm , m = 1, 2, 3, .... If Ĥ1 has p
bound states, then p Hamiltonians can be generated, and the m−th Hamiltonian Ĥm has the
same spectrum of Ĥ1 , except the first m − 1 levels of Ĥ1 that are obviously missing.
By repeated use of the shape invariance condition (66):
m−1
d2 X
(68) Ĥm = − 2 + V̂1 (x; am ) + R(ak ) ,
dx
k=1
where
m−1 times
z }| {
(69) am = f m−1 (a1 ) ≡ f (f (...f (a1 )...)) ,
i.e. f applied recursively m − 1 times. From Eqs. (66) and (68) the spectrum of Ĥm+1 is:
m
d2 X
(70) Ĥm+1 = − 2 + V1 (x; am+1 ) + R(ak )
dx k=1
m−1
d2 X
(71) = − + V 2 (x; a m ) + R(ak ) .
dx2 k=1
Ĥm and Ĥm+1 are SUSY partners, so they have identical bound state spectra except for the
ground-state of Ĥm of energy
m−1
X
(72) E0m = R(ak ) .
k=1
Going back from Ĥm to Ĥm−1 and so on, eventually Ĥ1 is reached, whose ground-state energy
is 0, and whose mth excited state level coincides with the ground state of the mth Hamiltonian
Ĥm . The complete Ĥ1 spectrum is thus
m
X
(1) (1)
(73) Em (a1 ) = R(ak ) , E0 (a1 ) = 0 .
k=1
Beside the eigenvalues, also the eigenstates can be obtained in algebraic form. Since the
operators Â and Â† link the eigenfunctions of the same energy for partner Hamiltonians, starting
(1)
from the Ĥm ground-state eigenfunction ϕ0 (x; am ), following the hierarchy chain and exploiting
Eq. (65), the nth un-normalized energy eigenfunction can be written as
(1)
(74) ϕ(1) † † †
n (x; a1 ) ∝ Â (x; a1 )Â (x; a2 )... Â (x; an )ϕ0 (x; an+1 ) .
Finally, Eq. (74) resembles the ladder representation of the harmonic oscillator, but there
(1)
are important differences. First of all the excited state ϕn (x; a1 ) is generated starting from
the ground state of a different but related potential problem. Secondly, the ladder operators
involved depend explicitly on the parameters, and through the recursive relation (69) they
depend explicitly also on the number of recursions, thus on the step number.
2.5.3. Application to the Morse problem. The SUSY formalism allows us to express
any shape-invariant potential problem in a form at the same time elegant and useful. In a
sense it leads to a generalization of the ladder operator methods commonly employed to treat
algebraically the harmonic oscillator problem, and also of the various exact and perturbative
method based on that expansion.
The Morse potential satisfies the condition (66), i.e. it is a SIP. It is thus exactly solvable,
but the SUSY approach adds the possibility of an algebraic treatment. We start applying the
formalism of previous section to the Morse potential. The SUSY Morse superpotential has
explicitly the form
(75) W (x) = C − De−α(x−x0 ) ,
that using Eq. (55), (58), provides for V1 (x), V2 (x) the explicit form

2 2 −2α(x−x0 ) α~
(76) V1 (x; C) = C + D e − 2D C + √ e−α(x−x0 )
2 2µ

2 2 −2α(x−x0 ) α~
(77) V2 (x; C) = C + D e − 2D C − √ e−α(x−x0 ) .
2 2µ
The SIP condition (66) is satisfied

(78) V2 (x; a1 ) = V1 (x; a2 ) + R(a1 ) ,
provided that
α~
(79) a1 = C , a2 = f (a1 ) = C − √ , R(a2 ) = a21 − a22 .
2µ
Applying Eq. (73), the mth eigenvalue of Morse Ĥ1 becomes
2
(1) 2 2 2 α~
(80) Em = a0 − am−1 = C − C − m √ .
2µ
Now, the potential V1 (x; a2 ) of Eq. (76) equals the Morse potential, except for a constant shift,
if
p p α~
(81) D = V0 , C = V0 − √ .
2 2µ
Substituting these values of the parameters C and D in the previous equations allows us to
obtain algebraically all Morse eigenvalues (Eq. (29)) and eigenstates (Eq. (26)). In particular,
the superpotential entering the operators A and A† (54) is
p α~ p
(82) W (x) = V0 − √ − V0 e−α(x−x0 ) ,
2 2µ
and the first two SUSY partner potentials V1 (x), V2 (x) (76)-(77) become
p
−2α(x−x0 ) −α(x−x0 ) α~
(83) V1 (x) = V0 [e − 2e ]+ V0 − √ ,
2 2µ
p p α~
p
α~

−2α(x−x0 ) −α(x−x0 )
(84) V2 (x) = V0 e − 2 V0 V0 − 3 √ e + V0 − √ .
2 2µ 2 2µ
2.5.4. Morse factorized Hamiltonian. We follow now a slightly different derivation of

Morse Hamiltonian (22) in the factorized form (53). We introduce the operators Â and Â† but,
in view of subsequent manipulations, following Ref. [14], we explicitly incorporate a parametric
dependence, as follows:
ŷ i
(85) Â(q) = q Iˆ −
+ p̂x ,
2 ~α
ŷ i
Â† (q) = q Iˆ − − p̂x ,
2 ~α
parameterized by the real quantity q. The operators (85) are characterized by the commutation
relations
(86) [Â(q), Â† (q 0 )] = (q + q 0 )I − (Â(q) + Â† (q 0 )), [Â(q), Â(q 0 )] = [Â† (q), Â† (q 0 )] = 0 .
The Morse Hamiltonian (22) expressed in factorized form is
~2 α 2 †
(87) ĤM = [Â (s)Â(s) − s2 ] ,
2µ
in terms of the operators of Eq. (85) with q = s, where s is defined in Eq. (24): the formal
analogy with the harmonic-oscillator algebra is evident. Define |Ψq0 i as the only state annihi-
lated by Â(q): Â(q)|Ψq0 i = 0. For q = s, |Ψq0 i coincides with the ground state of the Morse
Hamiltonian, Eq. (26). The other excited states can be obtained algebraically (see Eq. (74)) as
n
Y
(88) |Ψsn i = Fn Â† (s − n + m)|Ψs−n
0 i ,
m=1
where Fn are normalization constants.

The general considerations of the previous section regarding SIP applies to the Morse prob-
lem: equation (88) is reminiscent of the ladder representation of the harmonic oscillator but
(i) the excited state |Ψsn i is generated starting from the ground state of a different but related
Morse problem, and (ii) the ladder operators involved depend explicitly on the step number.
Until now this is only an elegant derivation of the solution of the Schrödinger equation for
Morse potential, completely equivalent to Morse original solution [24]. However, the SUSY
algebraic picture is particularly rich, and allows further manipulation in a natural way. The
simple dependence of the operators Â, Â† on the parameter q can be further exploited: different
choices of the step number q-dependency generate new algebraic families of wavefunctions,
which do not coincide with Morse eigenstates, but have similar algebraic properties.
In particular, setting q = σ + n, starting from the ket |φσ0 i ≡ |Ψs0 i annihilated by Â(σ), the
following family of wavefunctions is generated:
n
Y
(89) |φσn i = Gσn Â† (σ − 1 + m)|φσ0 i ,
m=1
where Gσn are appropriate normalization constants. By comparing Eq. (85) with Eq. (44), the
following relations emerge:
(90) Â(σ + n)|φσn i = K̂− |φσn i ,
Â† (σ + n)|φσn i = K̂+ |φσn i .
This means that, for this special (n-dependent) choice q = σ + n, the Â, Â† operators assume
the structure of the family of ladder operators (85), and the states (89) coincide with those
of Eq. (43). On the other hand, from Eq. (86) it can be seen that the algebraic structure
derived by the commutation properties of the Â and Â† operators is invariant with respect to
q-parameter translation, so the algebraic properties are preserved. These SUSY considerations
allow us to see the complete basis of Eq. (43) as a natural algebraic modification of the Morse
basis, with the nontrivial advantage that all matrix element of operators expressed as functions
of Â(q) and Â† (q) (see Eqs. (49) and (50)) can be computed by purely algebraic means. We
will take advantage of this feature in the next section to evaluate the matrix elements of all
operators involved in the expansion (33).
2.5.5. QNSB matrix elements. The first step to construct an effective exact diagonal-
ization method is the derivation of matrix elements of the Hamiltonian operator (22) and of the
relevant operators. The expansion of the correction potential in powers of v(x) = e −α(x̂−x0 ) − 1
(Eq. (23)) requires the knowledge of the matrix elements of the operator e−α(x̂−x0 ) ; once these
are available, all terms in expansion (33) of Vd (x), in the basis (43) of quasi number states, can
be obtained explicitly. In terms of the operators Â(q), Â† (q) defined in Eq. (85), the Morse
variable ŷ multiplication operator and the momentum operator p̂x can be written as:
h i
(91) ŷ = (2s + 1)e −α(x̂−x0 ) ˆ †
= 2sI − Â (s) + Â(s) ,
~α h †
i
(92) p̂x = Â(s) − Â (s) .
2i
The Â(s), Â† (s) matrix elements are computed explicitly by observing that
(93) Â(s) = Â(σ + n) + (s − σ − n)Iˆ ,
(and analogously for Â† ):
(94) hφσm |Â(s)|φσn i = Cn δm,n−1 + (s − σ − n)δm,n ,
hφσm |Â† (s)|φσn i = Cn+1 δm,n+1 + (s − σ − n)δm,n .
Based on the algebraic relations above, it is straightforward to obtain the matrix elements
of all relevant operators. In particular, the matrix elements of ĤM are
1
σ σ 2V0 2 1 2
hφm |ĤM |φn i = ~α [Cm − s2 + (m − s + σ)2 ]δm,n
µ (2s + 1)
(95) + (s − σ − n)Cm δm,n+1 + (s − σ − m)Cn δm+1,n .
In the quasi number states basis (QNSB) the Hamiltonian ĤM is therefore tridiagonal. In
practice, this is a minor drawback with respect to the basis (26) of energy eigenstates, since
tridiagonal matrices are diagonalized extremely quickly and corrections in Vd (x) could make
the Hamiltonian matrix non-diagonal anyway.
We are now ready to choose the value of the σ parameter: from Eq. (95) it is evident that
if σ − s equals a positive integer r, then the off-diagonal matrix element hr|ĤM |r + 1i vanishes:
σ determines the splitting of the Hamiltonian matrix into two blocks. It is natural to set the
integer r to the number [s] of Morse bound states, i.e.
(96) σ = s − [s] ,
so that 0 < σ ≤ 1. Under this assumption the Hilbert space H decomposes into the direct sum
of a ([s] + 1)-dimensional H− space (Morse bound states (26)), and an infinite dimensional H+
space (Morse continuum spectrum) [14, 35]:
(97) H = H− ⊕ H+ .
In the following we stick to the assumption (96).
In the same way of Eq. (95), the matrix elements of e−α(x̂−x0 ) are obtained from Eq. (91):
−Cn δm,n−1 − Cm δm,n+1 + 2(s − [s] + n)δm,n
(98) hφσm |e−α(x̂−x0 ) |φσn i = .
2s + 1
The matrix of e−α(x̂−x0 ) is also tridiagonal. Based on Eq. (98), all matrix elements of every
term in the expansion (33) can be computed exactly. For the reader’s convenience, explicit
expressions of the nonzero matrix elements of the hφn |[v̂(x)]i |φm i for i = 3 to 6 are collected in
Appendix 9.4.
In summary, the following extremely useful properties characterize the QNSB (43) with the
choice σ = s − [s]:
(1) the first [s] + 1 states generate the same Hilbert subspace as the Morse basis of bound
states;
(2) more states can be included simply, with a systematic improvement of the basis com-
pleteness;
(3) if extended to all n ≥ 0, it becomes a complete orthonormal basis of L2 [(0, ∞), dy/y],
allowing to expand the bound solutions of an arbitrary 1-dimensional problem, with
systematically improvable accuracy;
(4) most importantly, it allows to express the matrix elements of all relevant operators in
an algebraic form.
3 3 3
2 2 2
1 1 1
Ψ0(x) Ψ4(x) Ψ8(x)
0 0 0
-1 -1 -1
-2 -2 -2
0 2 4 6 0 2 4 6 0 2 4 6
αx αx αx
Figure 2.10. Morse ground-state wavefunction (left), 5th state wavefunction

(center), and highest bound state wavefunction (right), for the Morse potential
(α = 4, V0 = 625, x0 = 1, i.e. s = 8.34).
In particular, the matrix relative to a term [v̂(x)]i is (2i + 1)-band diagonal, so that the finite
expansion Vd and thus Ĥ is represented by an extremely sparse matrix, which is even a band
matrix.1 We propose therefore the QNSB as a convenient complete orthonormal system to
study the general problem of a molecular oscillator, in a way similar to what is usually done
in harmonic expansion methods, but with the advantage of including anharmonicity from the
beginning.
In Sect. 2.5.7 we develop and apply in detail this method to a couple of examples, and in
Chap. 4 we apply it to a couple of realistic potentials. In Chap. 5 we instead consider an even
more general situation. In the following section a comparison of Morse and QNSB wavefunction
is carried out.
2.5.6. Morse and QNSB wavefunctions. The bound states of Morse 1D Schrödinger
equation (26) depend on the same three parameters α, V0 , x0 which completely define the Morse
potential:
(99) Ψn (x) = Nn e−y(x)/2 y(x)s−n Ln2(s−n) (y(x)) , y(x) = (2s + 1) e−α(x−x0 ) .
The eigenfunctions (26) depend on the parameter x0 and α explicitly, but the V0 dependence
occurs only through the auxiliary parameter s (24), which also contains an α dependence:
√
2mV0 1
(100) s= − .
~α 2
If we take into consideration the explicit form of the Laguerre polynomials
ez z −µ dn µ+n −z
(101) Lµn (z) = (z e ) ,
n! dz n
we see that the general form of the nth bound state is:
(102) Ψn (y) ∝ e−y/2 y s−n P ol[y, n] ,
where P ol[y, n] indicates a polynomial of degree n in the y variable. Every eigenfunction is a
polynomial of degree n multiplied by y s−n e−y/2 . Figure 2.10 shows the behavior of the ground-
state wavefunction, the n = 4 and the highest excited state n = 8 for a Morse potential with
α = 4, V0 = 625, x0 = 1 (s = 8.34).
1 Like on the Morse basis, also in QNSB matrix elements of v(x)t for real t may be obtained analytically
in closed form.
3 3 3
2 2 2
1 1 1
σ σ σ
φ 0(x) φ 4(x) φ 8(x)
0 0 0
-1 -1 -1
-2 -2 -2
0 2 4 6 0 2 4 6 0 2 4 6
αx αx αx
Figure 2.11. QNSB 0th wavefunction (left), 5th wavefunction (center), and 8th
state wavefunction (right), for s = 8.34, compatible with a Morse potential char-
acterized by α = 4, V0 = 625, x0 = 1.
We can apply the same short analysis to the QNSB states of Eq. (43) labeled by a real
parameter σ > 0. The general form of the nth wavefunction is
(103) φσn (y) ∝ e−y/2 y σ P ol[y, n] ,
These wavefunctions, shown in Fig. 2.11, have a shape different from Morse eigenstates. Nonethe-
less, they are normalizable functions, and for the special choice σ = s − [s] of Eq. (96) the first
[s]+1 wavefunctions span the same subspace generated by the bound states of a Morse potential.
In this case, an explicit basis transformation relation is available:
n [s]+m−n
X X
i [s] + m − n
(104) |Ψn i = Nn C1 (n, 2(s−n), m) (−1) i!C2 (i) C3 (i, [s]+m−n)|φσi i ,
m=0 i=0
i
where
(n + k)!
(105) C1 (n, k, m) = (−1)m ,
(n − m)!(k + m)!m!
r
Γ[2σ + n]
(106) C2 (n) = ,
n!
and
n−1
(107) C3 (k, n) = Πj=k (2σ + j) ,
and Nn is defined in Eq. (28).
2.5.7. Numerical experiments. We come back here to the original purpose of the pre-
vious analysis: we want to verify if it is possible to overcome the intrinsic limitations of Morse
bound state basis without renouncing to the exactness and simplicity of the original solution.
We already claimed that using different approaches it is possible to identify a complete or-
thonormal system in the Hilbert space that allows to keep some degree of simplicity when
applied to distorted Morse Hamiltonians. As an additional (computationally) useful feature,
we also obtained that in spite of a slightly increased complexity in the treatment of Morse
problem (diagonal to tridiagonal matrix pure Morse Hamiltonian matrix), Hamiltonians built
according to the expansion (33) give rise to sparse matrices. It is now important to check the
validity of these claims, with some numerical experimentation.
Our first claim, is that the QNSB basis is exactly equivalent to Morse basis for the pure
Morse problem. For clarity, we redo the computation of Section 2.3.1, and find that the results
state Morse exact QNSB diagonalization

0 -556.28932188134 -556.28932188134
1 -430.86796564403 -430.86796564403
2 -321.44660940672 -321.44660940672
3 -228.02525316942 -228.02525316942
4 -150.60389693211 -150.60389693211
5 -89.182540694797 -89.182540694797
6 -43.761184457488 -43.761184457488
7 -14.339828220178 -14.339828220178
8 -0.918471982869 -0.918471982869
Table 2.8. Morse potential (α = 4, x0 = 1, V0 = 625, µ = 1, ~ = 1):
comparison of exact results and diagonalization of tridiagonal QNSB. Due to the
exact Hilbert-space decomposition (Eq. (97)), the results of diagonalization do
not depend on basis size, as long as it is greater than or equal to the number of
bound states.
state Morse basis Finite Differences QNSB (Ns = 30)

0 -562.48167 -562.48171 -562.48171
1 -436.83891 -436.83911 -436.83909
2 -326.99775 -326.99835 -326.99831
3 -232.98542 -232.98689 -232.98680
4 -154.82932 -154.83183 -154.83169
5 -92.557068 -92.56001 -92.55984
6 -46.187278 -46.19805 -46.19783
7 -15.728927 -15.77248 -15.77051
8 -1.25328 -1.30975 -1.28830
Table 2.9. Eigenvalues of the Schrödinger problem associated to the potential
of Eq. (108), with ` = 0.01. The Morse-basis diagonalization and the diagonal-
ization in a QNSB of Ns = 30 states are compared to the exact finite-differences
evaluation. The maximum discrepancy is ∆30 = 0.0034%.
coincides exactly for both choices of the basis. As can be seen immediately from Table 2.8, the
two methods are exactly equivalent for pure Morse potential, thus no approximation has been
introduced in the basis change.
We apply this method to the distorted Morse potential of Eq. (38) (Fig. 2.3):
(108) V (x) = VM (x) + a4 [v(x)]4 − V0 (1 + a4 ) .
for different values of the parameter a4 = `V0 , namely ` = 0.01, 0.02, 0.2, the same used in
Section 2.3.1.
The results of Tables 2.9 and 2.10 are encouraging: for small absolute values of the correction
strength R, if only a number of states equal to the number of bound states of the Morse potential
are included, the results are exactly equal to those obtained with Morse basis. The inclusion
of additional states in the QNSB improves the situation systematically in a variational fashion.
The greater qualitative success emerges instead from Table 2.11: when the QNSB contains
enough states, the additional bound state supported by the expanded potential is recovered.
The intrinsic limitation due to the finiteness of Morse basis is overcome:
• the QNSB is complete, so in principle exact results can be obtained;
state Morse Finite Differences QNSB (Ns = 30)

0 -568.67421 -568.67434 -568.67434
1 -442.81102 -442.81174 -442.81172
2 -332.55274 -332.55498 -332.55493
3 -237.95392 -237.95952 -237.95943
4 -159.06989 -159.07953 -159.07939
5 -95.95658 -95.96810 -95.96792
6 -48.63179 -48.67758 -48.67721
7 -17.08427 -17.25979 -17.25350
8 -1.54978 -1.76627 -1.68985
of Eq. (108), with ` = 0.02. The Morse-basis diagonalization and the diagonal-
ization in a QNSB of Ns = 30 states are compared to the exact finite-differences
evaluation. The maximum discrepancy is ∆30 = 0.012%.
state Morse basis Finite Differences QNSB (Ns = 30) QNSB (Ns = 60)
0 -680.16875 -680.17933 -680.17933 -680.17933
1 -550.49379 -550.54853 -550.54851 -550.54851
2 -433.12127 -433.31322 -433.31317 -433.31317
3 -328.53271 -329.09825 -329.09812 -329.09812
4 -237.48130 -238.42432 -238.42411 -238.42411
5 -159.92613 -161.74248 -161.74217 -161.74218
6 -92.21015 -99.45641 -99.45591 -99.456070
7 -37.12633 -51.93820 -51.93482 -51.93787
8 -4.86595 -19.54009 -19.49612 -19.53914
9 (missing) -2.60430 -2.31938 -2.53573
of Eq. (108), with ` = 0.20. The Morse-basis diagonalization and the diago-
nalization in two QNSB bases of Ns = 30 and 60 states are compared to the
exact finite-differences evaluation. The maximum discrepancies are, respectively,
∆30 = 0.046% and ∆60 = 0.011%.
• the QNSB preserves the algebraic structure of Morse potential problem, it is in a

certain sense a natural choice. As a consequence completeness is introduced keeping
the complexity of the problem manageable.
It has been shown that the QNSB method provides the correct eigenvalues for distorted
problems, at least of the simplest form. The completeness of QNSB can be verified also in a
more direct way by considering the projection of exact solutions on the subspace generated by
QNSB. As in Sect. 2.3.1, we compute the squared projection
N
X s −1
(109) |hφσj |ϕk i|2

j=0
of the k th ϕk eigenstate obtained via finite differences method on the subspace of a finite QNSB,
for different number Ns of states. As Figure 2.13 shows, the total projection of all bound states
approaches unity increasing the number Ns of states included in QNSB. We repeat the analysis
of Sect. 2.3.1 employing QNSB, for the same weights of the correction terms, but with different
8 9 9
100 9 100 8 8
7 7 8 7 7
6 6 6 6
0 0
5 5 5 5
-100 4 4 -100 4 4
-200 3 3
Energy
-200
Energy
3 3
R = 0.20
-300 2 2 -300 2 2
R = 0.20
-400 -400 1 1
1 1
-500 -500
0 0 0 0
-600 -600
-700 -700
0 1 2 3 0 1 2 3
x/x0 x/x0
Figure 2.12. Potential V (x) = VM (x) + 0.2 V0 [v(x)]4 − 1.2V0 . and eigenvalues
computed by means of a numerical diagonalization of the Hamiltonian repre-
sentation (dashed lines) in the Morse basis (left), and on the Ns = 30 QNSB
(right), both compared to the exact finite-differences eigenvalues (solid lines).
The highest bound state missing in the Morse diagonalization is recovered with
QNSB.
0.9
Projection
Ns = 9 states
0.8
Ns = 11 states
0.7
Ns = 18 states
0.6
0 1 2 3 4 5 6 7 8 9 10 11 12
k
PNs −1 σ 2 th
Figure 2.13. The projection j=0 |hφj |ϕk i| of the k eigenstate on the
subspace spanned by the QNS basis, for increasing relative importance R = 0.5
of the deformation Vd (x) = [v(x)]4 added to a Morse potential characterized by
[s] + 1 = 9 bound states, with different number Ns of states included in QNSB
(~ = 1, µ = 1, s = 8.34).
number of basis states. If we use [s] + 1 states, the exact number of Morse bound states,
we obtain exactly the same results of Sect. 2.3.1. Including additional QNS, the projection
tends to approach unity, as expected: additional QNS allows to describe the mixing to the
subspace orthogonal to Morse basis. Table 2.12 shows the projection (109) for the three highest
eigenstates for the potential Vd (x) = 1.0 [v(x)]4 obtained via a finite difference method, on a
QNSB of 30 elements.
As a quantitative measure of the error of the algebraic method, we apply here the worst-case
estimator
1
(110) ∆Ns := max |Ekalg − Ekex | ,
V0 k
state R = 0.01 R = 0.02 R = 0.20 R = 0.50

7 0.999995 0.999987 0.999999 0.999928
8 0.99849 0.996107 0.999933 0.998661
9 ———— ———— 0.991135 0.93898
Table 2.12. The projection of the near dissociation exact eigenstates on the
Ns = 30 QNSB. The projections of the first seven lowest lying states are extremely
close to unity.
0.5 R=1
Energy [V0]
0
13 13
12 12
2 2
-0.5 11 11
1 1
0 0 10 10
-1
1 5 9
αx
Figure 2.14. The Morse potential (dashed) and total potential with Vd (x) =
1.0 V0 [v(x)]4 (solid). The solid and dashed horizontal lines represent respec-
tively the exact and the variational eigenenergies obtained by diagonalization
on a QNSB of Ns = 30 states.
which for Vd (x) = 0.20 V0 [v(x)]4 equals ∆9 = 2.35% for the calculation based on the Morse
basis, or equivalently for the QNSB with Ns = 9. With at least Ns = 11 states included in the
QNSB, the bound states reach the correct number of 9 and the ∆Ns diminishes fairly rapidly
with increasing basis size: ∆20 = 0.051%, ∆30 = 0.028%, ∆40 = 0.019%.
The method provides accurate results even for potentials significantly distorted with respect
to the pure Morse. For the same potential expansion discussed above but with R = 1, i.e.
Vd (x) = 1.0 V0 [v(x)]4 , it produces the correct 14 bound states with an excellent ∆30 = 0.043%
(Table 2.13). Comparison with Table 2.5 shows the huge improvement in accuracy. Figure 2.14
shows the accuracy of the results.
Similarly good convergency emerges for a higher-degree simple distorted potential. Figure
2.15 and Table 2.14 refers to Vd (x) = 1.0 V0 [v(x)]6 . Using the same Ns the highest-energy
bound state is missing. Increasing Ns to 40 basis states, convergence of all the bound states is
achieved, even if a bit less satisfactory than the [v(x)]4 case, but the accuracy is still very high.
The results for the same potential expansion treated in the Morse basis are collected in Table
2.6 of Sect. 2.3.1.
A very severe distortion is represented by the potential, of Fig. 2.7 and Eq. (42) , already
analyzed with the Morse basis in Sec. 2.3.1. For this R = 8 correction potential, Morse basis did
not reproduce the two highest eigenvalues, and the accuracy was particularly poor (see Table
2.7). Even for this potential, the 11 bound states are obtained with extremely good accuracy,
the estimator of Eq. (110) giving of ∆30 = 0.11%, ∆40 = 0.011% (Table 2.15).
state Finite Differences QNSB (Ns = 30)

0 -551.353402 -551.353401 (5·10−7 %)
1 -407.916392 -407.916365 (5·10−6 %)
2 -266.824755 -266.824643 (2·10−5 %)
3 -131.411551 -131.411249 (5·10−5 %)
4 -3.7563923 -3.755760 (1·10−4 %)
5 114.663886 114.665005 (2·10−4 %)
6 222.712190 222.713898 (3·10−4 %)
7 319.462011 319.464754 (4·10−4 %)
8 404.124691 404.140228 (0.0025 %)
9 476.004188 476.089681 (0.014 %)
10 534.466198 534.692377 (0.036 %)
11 578.915461 579.189141 (0.044 %)
12 608.777324 608.933360 (0.025 %)
13 623.481379 623.588612 (0.017 %)
1, V0 = 625, µ = 1, ~ = 1): comparison of results of diagonalization of QNSB
(Ns = 30) and finite differences method. In parenthesis, the discrepancy in
percentage of V0 .
0.5 R=1
Energy [V0]
0
14 14
13 13
2 2
-0.5 12 12
1 1
0 0 11 11
-1
1 5 9
αx
Figure 2.15. The Morse potential (dashed) and potential expansion with
Vd (x) = 1.0 V0 [v(x)]6 (solid). The solid and dashed horizontal lines represent
respectively the exact and the variational eigenenergies obtained by diagonaliza-
tion on a QNSB of Ns = 40 states.
The number of states Ns necessary to obtain good accuracy throughout the whole spectrum
is an important issue in real application. There is no general answer, as this obviously depends
on the particular expansion. Our tests indicate that a ∆Ns within a small fraction of a per
cent is normally realized by including a number Ns of states about twice the number of bound
states. Occasionally, when high powers of v(x) are included in Vd (x) with substantial weight,
a larger Ns could be required for the same accuracy. Another difficult situation occurs when
the highest energy bound state happens to be very close to dissociation. In such cases the
convergence of the corresponding eigenfunction and eigenenergy can be slow, even if the weight

0 -554.871319 -554.871317 (4·10−7 %)
1 -422.790334 -422.790314 (3·10−6 %)
2 -296.419868 -296.419792 (1·10−5 %)
3 -174.253720 -174.253511 (3·10−5 %)
4 -56.865154 -56.864700 (7·10−5 %)
5 54.665764 54.666590 (1·10−4 %)
6 159.197037 159.198358 (2·10−4 %)
7 255.638778 255.640943 (3·10−4 %)
8 342.971475 342.974883 (5·10−4 %)
9 420.236370 420.242070 (9·10−4 %)
10 486.518288 486.595714 (0.012%)
11 540.926674 541.335782 (0.066%)
12 582.575888 583.416749 (0.13%)
13 610.564016 611.299017 (0.12%)
14 623.948568 624.139615 (0.031%)
1, V0 = 625, µ = 1, ~ = 1): comparison of results of diagonalization of QNSB
(Ns = 40) and finite differences method. In brackets the difference in percent of
V0 .

0 -567.687408 -567.687405 (5·10−7 %)
1 -493.327792 -493.327785 (1·10−6 %)
2 -415.473333 -415.473314 (3·10−6 %)
3 -337.808643 -337.808592 (8·10−6 %)
4 -263.617201 -263.616567 (1·10−4 %)
5 -195.325081 -195.323579 (2·10−4 %)
6 -134.836473 -134.820059 (0.0026%)
7 -83.766799 -83.530830 (0.038%)
8 -43.578937 -42.906055 (0.11%)
9 -15.678600 -15.135579 (0.087%)
10 -1.498858 -1.338373 (0.026%)
P
Table 2.15. Potential expansion with Vd (x) = 2.0 V0 6i=3 [v(x)]i , (α = 4, x0 =
1, V0 = 625, µ = 1, ~ = 1): comparison of results of QNSB (Ns = 30) and finite
differences method. In brackets the difference in percent of V0 .
R of the correction potential is small. Figure 2.16 illustrates this point clearly: for R = 0.5,
the highest bound state k = 11 is located at energy −0.00022 V0 , thus its wavefunction is
exceedingly extended, and its projection on the truncated QNSB grows extremely slowly with
Ns . In contrast, for R = 1, thus a much larger distortion, as the energy of the highest level
k = 13 is −0.0023 V0 , convergence is much more rapid. In general, substantially fewer states
can be included if only the low-lying levels are addressed, and on the contrary, Ns must be
substantially increased to reach convergence of the topmost bound state especially when it
happens to be located energetically very close to the continuum, thus associated to a very
extended wavefunction. We will come back to this problem in Chap. 5.
1 1
k = 11
k = 12
0.9 k = 13
0.9 K=9
k = 10
Projection
Projection
k = 11
0.8
0.8
0.7
4 4
0.7 VM(x) + 0.5 [v(x)] VM(x) + 1.0 [v(x)]
0.6
0.6 0.5
0 20 40 60 80 0 20 40 60 80
Ns Ns
P s −1 σ
Figure 2.16. The projection N 2
j=0 |hφj |ϕk i| of the three highest eigenstates
on the subspace spanned by the QNSB basis, for weight R = 0.5 (left) and
R = 1.0 (right) of the deformation Vd (x) = R V0 v(x)4 added to a Morse potential
characterized by [s] + 1 = 9 bound states, with different number Ns included in
QNSB (~ = 1, µ = 1, s = 8.34).
We have checked the accuracy of the QNSB-based method in treating molecular vibrational
potentials expressed as expansions of the form (34). If the potential expansion (33) is found to
be sufficiently flexible, we have an efficient tool to deal with fairly general anharmonic potential.
CHAPTER 3
The potential expansion
This section is devoted to an extensive analysis of the choice of the form of the potential
expansion. This is a general problem of functional approximation, but for the sake of brevity
and clarity we discuss it referring directly and explicitly to the Morse power expansion:
N
X max
(111) Vd (x) = ai [v(x)]i .

i=3
This work was inspired by the need to design a method to study vibrations in molecules
free from the known difficulties of harmonic-oscillator-based method (Chap. 1), and those of
the pure Morse oscillator (Chap. 2). In particular, the model potential should be physically
meaningful everywhere, i.e. V (x) must not produce unphysical behavior, for arbitrary chosen
truncation. This requirement alone leads to an approach “philosophically” different from the
Taylor series: the true potential must not be approximated only locally, but it is essential to
find a locally accurate approximation that preserves the global shape of the true potential.
The customary harmonic-oscillator-based methods (Eq. (1)) have indeed several advantages.
First, a standard procedure to approximate the potential is available. In the simplest situation
where the exact functional form of the real potential is assumed known, all the coefficients in
the expression (31) can be computed exactly by simple derivation. The real potential V real (x)
may be approximated by its Taylor series around the minimum x0 as:
N
X max
(112) V (x) = bi (x − x0 )i ,
i=2
i
where bi = i!1 dx
d
i V (x)|x0 .
Moreover, it is evident that only the k th derivative of the term (x − x0 )k in (31) is non-
vanishing in the minimum, and so each term in the expansion only contributes to the expansion
by fixing the k th derivative at the minimum to its exact value. Additional terms do not change
the value of the lower-degree coefficients: these benefits are consequences of an intrinsically local
method. These properties fail for a power expansion in a different variable. In the following we
limit ourselves to the problem of main interest to us, i.e. the expansion of Eq. (33).
In the minimum, each term of the form [v(x)]i = [e−α(x−x0 ) − 1]i has vanishing k th order
derivatives for 0 ≤ k < i, and non-vanishing ones for k ≥ i. Every term in the expansion (33)
contributes to determine the value of infinite local derivatives. For the sake of completeness,
the general form for the k th derivative, is
i
dk i k i
X
(113) k
[v(x)] = α (−1) cj e−ja(x−x0 ) [1 − e−α(x−x0 ) ]i−j ,
dx j=1
43
44 3. THE POTENTIAL EXPANSION
LJ potential Morse potential

ALJ
Well depth V (∞) − V (xmin ) 4
V0
1
Minimum position xmin 26 Σ x0
0 12Σ12 6Σ6 −2α(x−x0 ) −α(x−x0 )
First derivative V (x) ALJ (− x13 + x7 ) 2V0 α(−e +e )
00 2184Σ12 336Σ6 2 −2α(x−x0 ) −α(x−x0 )
Second derivative V (x) −ALJ ( x115 − x9 ) 2V0 α (2e −e )
000 2184Σ12 336Σ6 3 −3α(x−x0 ) −2α(x−x0 )
Third derivative V (x) −ALJ ( x115 − x9 ) 2V0 α (e − 4e )
2 A
2
Second derivative at minimum 9 · 2 3 Σ2LJ
2α V0
1 A
Third derivative at minimum −189 · 2 2 Σ3 LJ
−6α3 V0
Table 3.1. Expressions for the Lennard-Jones and Morse minimum position,
well depth and up to third-order derivatives in term of the respective parameters.
where the cj are numerical coefficients, whose explicit expression is derived in Appendix 9.3.
At the minimum:
i
dk [v(x)]i i k
X
(114) x=x0
= 0 for k < i, and (−1) α cj for k ≥ i .
dxk j=1
Notice that each term in the expansion (33) gives a non-vanishing contribution to the
dissociation limit. This means that the inclusion of additional terms in a lower-order previously
fitted expansion requires the coefficients of lower-order terms to change, in order to retrieve the
correct asymptotic behavior.
To discuss the problem of potential approximation in detail, let us concentrate on a specific
potential. It is useful to take for the potential to approximate an ideal potential, whose func-
tional form is analytically known: this will make some remarks clearer in the following. We
aim at investigating the consequences of different choices in the expansion (33) without losing
generality.
In the context of molecular 1D problem, we choose as reference potential the classical
Lennard-Jones (12,6):
" 6 #
12
Σ Σ
(115) VLJ (x) = ALJ − .
x x
The LJ potential has the characteristics of a realistic vibrational potential: one minimum, a
repulsive barrier at short distance x, and the correct asymptotic attractive behavior at small
and large displacement from equilibrium. VLJ is indeed widely used as prototype interatomic
potential in computations of liquid and disordered systems.
Table 3.1 collects the explicit form of the minimum position, the well depth and the three
lowest derivatives for LJ and Morse potentials in term of the respective parameters. They are
useful to determine the exact relations among the parameters emerging from the approximation
procedure.
To approximate the LJ potential with a pure Morse term, the local approach would suggest
to fix the three Morse parameters imposing the equalities of the minimum position
1
(116) x0 ≡ 2 6 Σ ,
of the second derivative at the minimum
2 ALJ
(117) 2α2 V0 ≡ 9 · 2 3 ,
Σ2
3.1. POTENTIAL FITTING SCHEMES 45
Morse
Energy [ALJ/4]
LJ
-0.5
-1
0 1 2 3
1/6
x /(2 Σ)
Figure 3.1. LJ and Morse potentials, sharing position of the minimum, second
derivative in the minimum, and well depth.
and the third derivative at the minimum

ALJ 1
(118) − 6α3 V0 = −189 · 2 2
.
Σ3
However, this choice yields a completely different dissociation threshold. We therefore prefer
to replace Eq. (118) with the equality of the well depth
ALJ
(119) V0 ≡ .
4
Under these assumptions, the analytic expressions for the Morse parameters in terms of the
LJ ones are:
1
(120) x0 = 2 6 , Σ
ALJ
(121) V0 = ,
4
3 · 25/6
(122) α = .
Σ
As Fig. 3.1 shows, the agreement is good only near the minimum. This is expected, because
LJ and Morse potential are the prototypes of two quite different approaches to dissociation.
3.1. Potential fitting schemes

Once we decide to go beyond the limitation of pure Morse potential and introduce the
correction potential Vd (x) as in Eq. (33), we must make sure that the chosen expansion is
meaningful. To avoid the problems that occur with the Taylor expansion [3], we must not
neglect an accurate analysis of the global behavior of the truncated expression. On the other
hand, this expansion is more complicated than the Taylor one, and there are no specialized tools
to produce an a priori answer, so we follow a cautious approach, that will lead to a detailed
understanding.
It is important to check the flexibility of the expansion (33): the subset of correction
potentials [v(x)]i identified as physically significant must be sufficient to describe accurately a
wide enough class of admissible 1D molecular potentials. If this was not the case, the model
would be simply a theoretical exercise, with little hope of applicability.
The traditional approach [40, 41] consists in exploiting the similarities between the Dunham
expansion for vibrational energy levels and the Morse energy spectrum to obtain relations
between Morse parameters α, V0 (or s) and traditional spectroscopic constants ωe , ωe xe . This
approach is best suited to start from spectroscopic data, but in our Hamiltonian approach it is
0
4
v(x) -1
6
v(x) -1
Morse
Energy [V0]
-0.5
v(x) even powers
-1
0 1 2 3
αx
Figure 3.2. Morse potential (solid) and (v(x)4,6 − 1) (dot-dashed).
not natural. Indeed, in Ref. [26] some remarks are made about the choice of what one could
call a “best” unperturbed Morse potential, and the authors there suggest to use the one which
fits the actual potential function in the region of the minimum, having the same location of the
minimum, and the same second and third derivatives there. We will show that this problem is
related to the form of the expansion chosen.
We start imposing strict limitations and thus to identify a subset of admissible functions as
correction potential. Next we will try to enlarge the subset to obtain greater flexibility without
renouncing to the global significance.
At first we follow Ref. [26] and impose the condition that the target and model potential
always share the location of the minimum, the values of second and third derivatives there, and
the well depth.
We start restricting the potential expansion to the form:
N
X max
(123) V (x) = VM (x) + Vd (x) , Vd (x) = a22i [v(x)]2i .

i=2
This is a very conservative choice. Each term in the expansion represents by itself a well-
behaved anharmonic potential (Fig. 3.2), characterized by a repulsive region, a dissociative
region, and one well defined minimum. The positive sign of all coefficients ensures that the
total potential is the sum of anharmonic potentials, thus necessarily a well behaved potential
itself.
As a second step we successively relax the constraint and employ a potential expansion of
the form
Nmax
X−1
(124) V (x) = VM (x) + Vd (x) , Vd (x) = a2i [v(x)]2i + a22Nmax [v(x)]2Nmax .
i=2
It still consists in an even-power expansion, and the only formal difference is that the sign of the
weights of the intermediate terms is no longer fixed. The coefficient of the highest order term
must be strictly positive to ensure the correct asymptotic behavior for x → −∞. Due to the
uncertain sign of the coefficients, the overall behavior of the total potential must be checked a
posteriori. In both cases the equality of second and third derivative affects only the pure Morse
term that consequently becomes completely specified, even in its depth. The explicit relations
to the LJ parameters are:

1
(125) x0 = 2 6 Σ ,
9
(126) V0 = ALJ ,
49
7 5/6
(127) α = 2 .
2Σ
Next, we subsequently check a PMO expansion, as in Eq. (32), where we should take care
that aNmax > 0. This expansion allows the use of odd power terms, and thus widens the class of
functions substantially. This expansion obviously requires an a posteriori check of consistency
of the resulting complete potential with the physical constraints. This provides a compromise
between the search for a good local agreement, the need to limit the total weight of correction
terms with respect to the pure Morse term, and the necessity of avoiding the use of v(x)3 terms
for reasons related to the perturbative approach employed in Ref. [26]. Note that the absence
of the cubic power makes the conditions on the derivatives affect only Morse term, even in this
case.
We tested the accuracy of the three expansions (123), (124) and (32) by fitting each of them
to a pointwise discretized LJ potential, to mimic the approach employed in a realistic situation,
where the adiabatic potential is known by points1, plus we impose the conditions on the second
and third derivatives exactly. The accuracy obtained is not encouraging, though many works
in literature [26, 27, 28, 29, 30] claim great results, especially for the PMO expansion of
(32). Fair accuracy requires expansions up to very high order, and the accuracy of the results is
not improving systematically nor particularly fast. Moreover, the controllability of the global
behavior of this expansion becomes a cumbersome task.
A more accurate analysis suggests that the main problem resides not only in the expansion
chosen, but in the difficulty to reconcile a global accuracy and local agreement in the important
region around the minimum. In all the previous models the request of perfect local agreement
(through second and third derivatives) determines completely the Morse parameters V0 , x0 and
even α. α, in particular, determines the global shape of the Morse term and of all v(x) powers.
In other words, through α, the requirement of local accuracy determines the global shape of all
the correction terms. This is a very strong limitation, that manifests itself heavily in the case of
a target potential such as the LJ one that shows a minimum-to-dissociation trend very different
from Morse (or exponential) one. The importance of this aspect is seen comparing the accuracy
of the fit of the previous expansion relaxing the requirements above. In particular, we drop the
request about the third derivative and retain the equality of the second derivative. This still
imposes some degree of local accuracy. For example, by applying Eq. (124) and Eq. (32) we
obtain the results of Table 3.2.
As can be seen from Table 3.2 and Fig. 3.3, dropping the constraint about the exactness
of the third derivative improves substantially the accuracy of the fit, though weakening the
agreement in the region near the minimum. In other words, the Morse parameter α has better
be fitted in order to improve the overall agreement. The third derivative is now no longer exact,
which is the price to pay in this rigid scheme.
This last drawback is solved by inserting the [v(x)]3 in the PMO expansion (32), thus
obtaining Eq. (33): in our scheme the PMO constraints are no more important, because the
QNSB allows to manage potentials in which the weight of correction terms is not small, as
shown in Sec. 2.5.1 and following. Moreover, in our model there is no particular reason to
1We fit approximately 1000 points in the interval between the two points xA,B (see Fig. 1.2) such that
VLJ (xA,B ) = 1%VLJ (x0 ). The fit is a standard minimization of the root mean square difference of the model
(e.g. Eq. (124)) and target (e.g. LJ Eq. (115)) potentials at the grid points.
Model χ
Eq. (123) Nmax = 4 0.054
nd rd
2 and 3 Eq. (124) Nmax = 6 0.043*
Eq. (32)Nmax = 12 0.019
Eq. (124) Nmax = 6 0.016
2nd Eq. (32) Nmax = 12 0.0025
Eq. (33) Nmax = 6 0.00042
p
2
Table 3.2. χ = χ quality of the fit of the expansion corresponding to the
equations in column 2. The first three rows refer to fits in which equalities involv-
ing second and third derivatives are imposed. Rows 4 and 5 refer to fits in which
only second derivative equality is imposed: the gain in accuracy is evident. The
last row refers to the expansion of Eq. (33), second and third derivatives exact,
and a low degree expansion: the accuracy improves substantially (∗ indicates a
rather poorly converged fit).
0 0 0
Eq. (123), Nmax = 4
Eq. (124), Nmax = 6
Energy [ALJ/4]
Energy [ALJ/4]
Energy [ALJ/4]
LJ LJ LJ
-0.5 -0.5 -0.5
VM(x)
c)
a) b)
-1 -1 -1
0 1 2 3 0 1 2 3 0 1 2 3
1/6 1/6 1/6
x /(2 Σ) x /(2 Σ) x /(2 Σ)
0 0 0
Eq (124), Nmax =12

Eq. (32), Nmax = 12
Energy [ALJ/4]
Energy [ALJ/4]
Eq. (32), Nmax = 6

Energy [ALJ/4]
LJ LJ LJ
-0.5 -0.5 -0.5
d) e) f)
-1 -1 -1
0 1 2 3 0 1 2 3 0 1 2 3
1/6 1/6
x /(2 Σ) x /(2 Σ) x [α]
Figure 3.3. a) LJ potential (solid) and Morse term sharing second and third
derivatives. b) Eq. (123) with Nmax = 4: higher order fit (up to order 8) do
not improve the agreement much. c) Eq. (124), up to Nmax = 6: x0 and first
and second derivative in x0 , well depths coincide. d) Eq. (32) up to Nmax = 12,
derivatives exact. e) Eq. (124), equality of the third derivative not required. f)
Eq. (32), condition on the third derivative dropped.
exclude a priori any power of v(x), so we include the third-power term. In turn this allows
to restore exact 3rd derivative. The equality of the third derivatives determines an equation
involving the two parameters α and a3 :
(128) − 6α3 (V0 + a3 ) = V 000 (x0 ) .
This choice does not fix rigidly the shape of the Morse term, leaving enough room for a globally
accurate fit. The last row of Table 3.2 refers to this choice: the improvement is substantial
even for a very low-order expansion, Nmax = 6.
0 0
Morse Morse
+ +
expansion (Nmax = 4)
Energy [ALJ/4]
Energy [ALJ/4]
expansion (Nmax = 6)
-0.5 LJ -0.5 LJ
-1 -1
0 1 2 3 0 1 2 3
1/6 1/6
x /(2 Σ) x /(2 Σ)
Figure 3.4. LJ potential (solid) and different fit functions: (34) with Nmax = 4
(left, 1 free parameter), and (34) with Nmax = 6 (right, 3 free parameters).
Parameters Nmax = 4 Nmax = 6

α 3.37587 2.418165
V0 2.507 4.886
a3 2.124 7.715
a4 0.617 6.017
a5 2.851
a6 0.6629
χ 0.0046 0.00042
Table 3.3. Optimal parameters of the best fit of expansion (33), with
Nmax = 4, 6. The last line reports the value of the root mean square χ.
This analysis is only partial, but indeed suggests that the criterion of local agreement often
fights against the overall global agreement. Moreover, the use of an expansion of the form (33)
is encouraged, because at the same time it allows greater flexibility and reduces the importance
of higher-order terms. This is evident in the case of LJ potential.
The reasons for the different capability of the various schemes may be partly disguised in
case of fitting of a target potential not analytically known. In this last case, in fact, it is not
really possible to strictly fix the local behavior, and the natural approach privileges a global
agreement, thus underestimating the role of exact derivatives.
As our final choice we employ the model potential (34). For comparison, we report the
corresponding fit for truncations at Nmax = 4 and Nmax = 6, with the conditions of equal
second and third derivatives at the minimum. The adjusted parameters are Nmax − 3. Fig. 3.4
shows the curve agreement. The quantitative estimate of the pointwise agreement of these fits
is reported in Table 3.3.
In both cases the fit leads to globally well-behaved curves, so we verify a posteriori that
the conditions we impose are consistent with the model. The agreement between the target
potential and our Morse expansion is very good already for Nmax = 4, and it improves by about
one order of magnitude for Nmax = 6.
In conclusion, the expansion (33) is very flexible: it can account naturally for substantially
distorted potentials such as those depicted in Fig. 2.12, 2.14, and even double-minima configu-
rations of the kind sketched in Fig. 3.5. Henceforth we will therefore use a model potential in
the form (33) to describe stretching potentials in the most general context.
0
(1,2,2,4,4)
(1,2,2,2,2)
(1,2,2,6,6)
(1,2,2,8,8)
(1,0,0,0,0)
Energy [V0]
-0.5 4-terms expansions
-1
0 1 2 3
αx
P6 i
Figure 3.5. Morse potential (solid) and V (x) = i=2 ai v(x) , with
(a3 , a4 , a5 , a6 ) = V0 (2, 2, 2, 2), (2, 2, 4, 4), (2, 2, 6, 6), and (2, 2, 8, 8) (dashed). No-
tice the appearance of a double well shape.
CHAPTER 4
Applications to diatomic systems
To explore the applicability of the theory outlined in Chaps. 2 and 3 to actual systems, we
consider initially the purely vibrational spectra of diatomic molecules (the results of the present
chapter were reported in Ref. [42]).
4.1. The Lennard-Jones potential

As a first example, we probe the ability of the Morse expansion (33) to describe accurately
the standard (12, 6) LJ potential (115), already sketched in Chap. 3, and apply the theory of
Chap. 2 to compute the vibrational spectrum . Like the Morse potential, VLJ is a model po-
tential, which however reproduces well the large-x behavior of the actual interatomic adiabatic
potential of, e.g., a dimer of noble gas atoms. Due to its power-law dependencies, VLJ can be
considered as a rather “tough” potential to approximate by a series of Morse-like terms.
As the functional form of the target function V (x) is assigned, it is straightforward to
extract a few exact properties, such as well depth and derivatives at the minimum, which can
be imposed to the approximating potential VM + Vd . Exact equality of these properties can be
used to fix a number of constraints among the Nmax + 1 parameters which determine VM + Vd .
The choice of the imposed exact equalities leaves some degree of arbitrariness to the fitting
scheme. It is possible to enhance global versus local accuracy, by privileging properties such as
the well depth, or rather a number of exact derivatives at the minimum. An extensive analysis
of this sort was applied in Chap. 3, and a similar discussion can be found in literature, in a
different but related context [26].
To start, we set Nmax = 2, i.e. Vd = 0 and analyze the pure Morse potential. By imposing
equal position of the minimum, equal well depth, and equal value of the second derivative at the
minimum, we fix all three Morse parameters, and obtain the dashed curve drawn in Fig. 4.1.
These conditions are the natural choice to ensure an overall agreement, which however is not
particularly good, as the LJ and Morse potential are the prototypes of two quite different
behaviors. As expected, the two potentials differ the most far from the equilibrium distance,
in the approach to the dissociation limit.
Nmax = 4 Nmax = 6
δRM S 0.0046 0.00043
α 0.10891 0.078138
V0 2.507 4.8698
a3 2.124 7.6676
a4 0.617 5.9625
a5 2.8206
a6 0.6560
Table 4.1. Optimized parameters in the expansion (33) with Nmax = 4 and 6,
fit to the reference LJ potential with Σ = 31, ALJ = 4 (so that in the well of unit
depth fit 10 bound states, for ~ = 1, µ = 1).
51
52 4. APPLICATIONS TO DIATOMIC SYSTEMS
B
0
A
Energy [ALJ/4]
Morse
-0.5 LJ
-1
1 2
1/6
x /(2 Σ)
Figure 4.1. The Morse potential (dashed) approximating the LJ potential

(solid), with the requirement of having equal position of the minimum, the same
second derivative at the minimum, and the same well depth. The points A and B
are characterized by V (x) = 1% V (x0 ), i.e. 99% of the dissociation energy from
the well bottom.
We then use the full expansion (33) truncated at Nmax = 4 and Nmax = 6. We impose the
following conditions: (i) same location of the minimum (fix the value of x0 ), (ii) same values
of second and third derivatives at the minimum (fix a relation between α, P V0 , and a relation
involving α and V0 and a3 , see Eq. (128)), (iii) same well depth (fix V0 + i (−1)i ai ). These 4
conditions reduce the number of adjustable parameters to Nmax −3. To determine these residual
free parameters, we minimize the RMS difference of V (x) and VM (x) + Vd (x) for approximately
103 equally spaced x-points between the points A, B of Fig. 4.1, where V (x) = 1% V (x0 ), i.e.
99% of the dissociation energy measured from the well bottom. The optimized parameters and
the corresponding RMS discrepancy of the fit are reported in Table 4.1. Note that in both
calculations the fitting potential is globally well-behaved, as guaranteed by aNmax > 0. The
agreement between the target potential VLJ and the approximating expansion VM + Vd is very
good already for Nmax = 4 (one fitted parameter only), and the RMS discrepancy improves by
about one order of magnitude for Nmax = 6 (three fitted parameters). Both fitted potentials
are visibly indistinguishable from VLJ , and have therefore not been drawn in Fig. 4.1. The fit
is essentially independent of the density of the mesh points, except for the placement of the cut
at the repulsive side. 99% of the dissociation energy is a reasonable compromise to address the
bound states.
This accurate fit of V (x) produces extremely good accord for the corresponding Schrödinger
eigenvalues, as shown in Table 4.2. A fairly accurate spectrum (∆40 = 0.46%) is obtained
already for Nmax = 4: the highest eigenvalue is missing but, due to its proximity to dissociation,
it is indeed very difficult to determine. The fit with Nmax = 6 recovers the missing eigenstate,
and improves substantially the accuracy of the eigenvalues (∆40 = 0.068%), determined as
usual by the state closest to dissociation. Note that the absolute accuracy of the eigenvalues is
of the same order as the accuracy of the underlying functional fit of the potential.
Recall the analogous result of a high-order Taylor expansion reported in Fig. 1.2 and Table
1.1: it is apparent that the traditional approach can only reach the accuracy of a few percent
on few low-lying levels, but fails completely starting already at moderately low overtones. In
contrast, even the rough approximation of the pure Morse term provides level estimates within
4.2. THE H2 MOLECULE 53
n LJ (exact) Nmax = 4 Nmax = 6

0 -0.88237 -0.88259 (-0.02%) -0.88240 (-0.002%)
1 -0.67488 -0.67592 (-0.10%) -0.67494 (-0.006%)
2 -0.50142 -0.50387 (-0.25%) -0.50146 (-0.004%)
3 -0.35948 -0.36333 (-0.39%) -0.35942 (0.005%)
4 -0.24637 -0.25098 (-0.46%) -0.24623 (0.014%)
5 -0.15927 -0.16343 (-0.42%) -0.15916 (0.011%)
6 -0.09514 -0.09751 (-0.24%) -0.09520 (-0.006%)
7 -0.05078 -0.05040 (0.04%) -0.05100 (-0.022%)
8 -0.02278 -0.01973 (0.30%) -0.02286 (-0.008%)
9 -0.00754 -0.00356 (0.40%) -0.00711 (0.043%)
10 -0.00126 (missing) -0.00058 (0.068 %)
Table 4.2. Bound-state energies for the LJ potential (Σ = 31, ALJ = 4)
computed by numeric finite-differences solution of the Schrödinger equation (units
of ~ = 1, µ = 1), compared to the eigenvalues of the problem associated to the
best fit VM (x) + Vd (x), solved by diagonalization in the QNSB including Ns = 40
states. All reported figures are significant. In brackets: the difference in percent
of the well depth Ed = ALJ /4.
few percent of the VLJ levels throughout the spectrum. The expanded potential yields a good
global description of the spectrum. It would be straightforward to apply the present method
to the recently determined adiabatic potential of the Ar2 dimer [43], but we prefer to consider
in detail a molecule whose PES is less close to the LJ function.
If we repeat the diagonalization of the Nmax = 4 and 6 approximation to the LJ potential,
but in the set of Morse bound states, rather than in the QNSB, we obtain essentially the same
figures as in Table 4.2, except for the least significant digit of a couple of levels of the Nmax = 4
expansion. The reason for this accord is that the Morse well in the expansion for Nmax = 4
is wide and deep enough to host 21 bound states, and for Nmax = 6 to host 40 bound states.
As the target LJ potential only has 11 bounds states, both Morse bases states are sufficiently
complete to reproduce accurately the correct behavior.
This makes us suspect that fewer than the Ns = 40 states included may suffice to deter-
mine these 11 bound states. Table 4.3 compares the eigenvalues computed for the Nmax = 4
expansion in three different bases each of 15 elements: the Morse basis, the QNSB and the
QNSB+ (introduced in Chap. 5). The numerically exact (finite-differences) spectrum of the
same potential is taken as reference. As can be seen, the Morse basis produces well converged
results for the lowest 7 bound states, while for the 8th one the agreement is less good, and two
bound states are missing. On the other hand, the QNSB with the same number of elements
yields poor agreement along the whole spectrum. A different choice of the parameters s = 15.9
and σ = 0.9 improves substantially the results: for the QNSB+ basis of 15 states, the estimator
∆Ns := E1d maxk |Ekalg − Ekex | is as low as 0.041%. Note however that s and σ in the QNSB+
have been optimised to minimize the discrepancy.
4.2. The H2 molecule

We acquire the accurate ab-initio determination of the H2 adiabatic potential of Refs. [23].
In our parametrization, we choose to fix both the well depth to the very highly accurate value
of Ed = 0.17447498 Eh (Eh indicates the Hartree energy, 27.2114 eV), and the minimum
position x0 = 1.4011 a0 (a0 indicates the Bohr radius, 0.529177·10−10 m) [44]. The other
n Fit LJ Nmax = 4 (reference) Morse (Ns = 15) QNSB (Ns = 15) QNSB+ (Ns = 15)
0 -0.88259 -0.88257 (0.002%) -0.88188 (0.071%) -0.88257 (0.002%)
1 -0.67592 -0.67592 (0.0001%) -0.67174 (0.418%) -0.67586 (0.006%)
2 -0.50387 -0.50386 (0.001%) -0.49365 (1.022%) -0.50373 (0.014%)
3 -0.36333 -0.36333 (0.0001%) -0.34764 (1.569%) -0.36312 (0.021%)
4 -0.25098 -0.25097 (0.001%) -0.23263 (1.835%) -0.25069 (0.029%)
5 -0.16343 -0.16342 (0.001%) -0.14547 (1.796%) -0.16306 (0.037%)
6 -0.09751 -0.09745 (0.006%) -0.08221 (1.530%) -0.09710 (0.041%)
7 -0.05040 -0.04610 (0.43%) -0.03907 (1.133%) -0.05002 (0.038%)
8 -0.01973 missing -0.01295 (0.678%) -0.01948 (0.025%)
9 -0.00357 missing -0.00089 (0.268%) -0.00346 (0.011%)
Table 4.3. Bound-state energies for the Nmax = 4 fit to LJ potential (de-
fined by the parameters listed in the appropriate column of Table 4.1) com-
puted by numeric finite-differences solution of the Schrödinger equation (units
of ~ = 1, µ = 1), compared to the eigenvalues of the same problem, solved by
diagonalization in the Morse basis. The QNSB and the QNSB+ (defined later
in Chapter 5), including Ns = 15 states. The pure Morse term allows 21 bound
states. All reported figures are significant. In brackets the difference in percent
of the well depth Ed = ALJ /4.
Morse Nmax = 4 Nmax = 12

δRM S [Eh ] 0.035 0.006 0.00003
δRM S [cm−1 ] 7760 1320 5.5
δRM S /Ed 0.20 0.03 0.0001
α 1.318 a−1
0 1.223 a −1
0 0.840657 a−1
0
V0 0.17447498 Eh 0.120 Eh 0.26147 Eh
a3 -0.030 Eh 0.09233 Eh
a4 0.024 Eh 0.06147 Eh
a5 0.03643 Eh
a6 -0.00995 Eh
a7 -0.02819 Eh
a8 0.04013 Eh
a9 0.02555 Eh
a10 -0.04034 Eh
a11 0.01243 Eh
a12 0.00025 Eh
Table 4.4. Best fit parameters for model potential (33), Nmax = 2, 4, 12.
Nmax − 1 parameters, including a constant setting the overall energy zero, are all adjusted by
minimization of the RMS deviation δRM S from the ab-initio points provided in Ref. [23]. The
mesh consists of all 169 available points ranging from 0.2 a0 to 18 a0 representing the effective
BO potential for H2 as computed in Ref. [23]. The fits include all available points. In order
to enhance the convergence in the bound-state energy range, the points above 1% V (x0 ) are
4.2. THE H2 MOLECULE 55
2 -0.05
Energy [Eh]
1.5 -0.1
1 -0.15
1 2 3
0.5
0
0 2 4 6 8
x [bohr]
Figure 4.2. Adiabatic potential for H2 , from Ref. [23] (dots). Best fit pure
Morse potential with α as the only adjusted parameter (dashed), and fit function
(33), with Nmax = 4 (dot-dashed) and Nmax = 12 (solid). Inset: detail of the
minimum region, where the small inaccuracy of the dot-dashed curve makes it
distinguishable from the solid curve.
weighted 1/9 of those inside the well region1. A first fit with the Morse potential term alone
yields a rough approximation of VH2 (δRM S = 0.035 Eh ), as shown in Fig. 4.2. We then fit the
expansion (33) with Nmax = 4 and 12, obtaining δRM S = 0.006 and 0.00003 Eh respectively.
The best-fit parameters are listed in Table 4.4. Figure 4.2 reports the resulting profiles.
The model Hamiltonian is solved on a QNSB including the first Ns = 60 states: the
computed levels are listed in Table 4.5. The obtained eigenenergies are to be compared with
those obtained by numerical integration of the Schrödinger problem of a third-order spline
going through the points of Ref. [23]. These reference energies (second column of Table 4.5)
differ from the levels obtained in Ref. [23] (also listed) in that they do not include non-adiabatic
corrections. The agreement improves systematically as Nmax increases, with a worst discrepancy
of 324 cm−1 for Nmax = 4, which reduces to 7 cm−1 for Nmax = 12. If the state closest to
dissociation is not considered, all other levels are computed within 2.4 cm−1 of their reference
values.
Interestingly the Nmax = 12 spectrum could equally well be computed on a QNSB of Ns = 30
states only, and the same cm−1 numbers are obtained. The reason is that the underlying Morse
term is so wide and deep that it could host 26 bound states, so that even the simple set of
Morse bound states would be almost complete for the bound states of the well at hand. The
Nmax = 4 calculation generates an extra spurious bound state whose precise energy location
converges only extremely slowly with Ns , while all other states require Ns = 60 to remain stable
to about one cm−1 . This is due to the narrow shallow Morse well underlying this expansion.
In such a kind of expansion, the QNSB is absolutely necessary: the Morse basis would produce
very poorly converged spectra.
1Complete exclusion of the points above 1% V (x0 ) produces better agreement of low-order fits, but makes
higher-order fits (Nmax > 8) unstable. On the other hand, enforcing equal accuracy to all points (no weights)
makes all fits much less accurate.
n Ref. [23] Spline Pure Morse Eq. (33) Nmax = 4 Eq. (33) Nmax = 12
cm−1 cm−1 cm−1 cm−1 cm−1
0 -36118 -36112 -35524 (2%) -36136 (-0.06%) -36113 (-0.002%)
1 -31957 -31948 -30298 (4%) -31963 (-0.04%) -31948 (-)
2 -28031 -28021 -25488 (7%) -27979 (0.1%) -28020 (0.001%)
3 -24336 -24324 -21093 (8%) -24203 (0.3%) -24324 (-)
4 -20868 -20854 -17114 (10%) -20651 (0.5%) -20856 (-0.004%)
5 -17626 -17612 -13550 (11%) -17340 (0.7%) -17614 (-0.005%)
6 -14612 -14598 -10402 (11%) -14282 (0.8%) -14599 (-0.004%)
7 -11830 -11815 -7670 (11%) -11491 (0.8%) -11815 (-)
8 -9287 -9272 -5353 (11%) -8979 (0.8%) -9271 (0.003%)
9 -6994 -6980 -3451 (9%) -6755 (0.6%) -6979 (0.003%)
10 -4967 -4955 -1965 (8%) -4829 (0.3%) -4955 (-)
11 -3230 -3220 -895 (6%) -3213 (0.02%) -3222 (-0.006%)
12 -1816 -1807 -240 (4%) -1915 (-0.3%) -1810 (-0.01%)
13 -766 -760 -1. (2%) -943 (-0.5%) -761 (-0.002%)
14 -145 -142 (missing) -308 (-0.4%) -135 (0.02%)
15 -7
Table 4.5. Comparison of the computed H2 bound-state energies: the spectrum
of Ref. [23] includes non-adiabatic corrections; when these corrections are left
out, and the bound states of the splined potential of Ref. [23] are computed
by numerical solution of the differential Schrödinger problem, the energies of the
second column are obtained. The successive columns report the eigenvalues of the
pure Morse term and of the expansion (33), fitted to the same adiabatic potential
of Ref. [23], and computed on a QNSB of Ns = 40 states. Discrepancies to the
energies in the second column, in percent of Ed , are reported in brackets: those
of less than 0.001% are replaced by a dash.
CHAPTER 5
Unlinking of basis and potential parameters
The QNSB exact-diagonalization method was introduced through Sec. 2.5.4 and Chap. 4
and analyzed in detail. In Chap. 2 we verified its ability to solve test potential expansions in
the Morse variable v(x) of Eq. (23); in Chap. 4 we applied it carefully to realistic diatomic
potentials.
The potential expansion (34) provides a good compromise between the ability in producing
globally accurate approximations of a typical 1D molecular vibrational potential, and the flexi-
bility necessary to reproduce the local details of the important minimum region. Moreover, the
risk that such expansion produces unphysical behavior is not ruled out, but the global behavior
of the expansion is relatively easy to assess. In general, the weight of the pure Morse term
in the best fit expansion often turns out to be comparatively small with respect to the other
terms, and the completeness of the QNSB (43) is therefore essential in allowing to apply the
method in such non-perturbative context.
Among the various complete basis of the form (43), QNSB is the one most strictly related to
the pure Morse potential. One may wonder if the QNSB is an optimal choice to treat a potential
significantly distorted away from the pure Morse potential. This aspect can be particularly
useful in a context where computational efficiency is important, typically in polyatomic molecule
computations [9, 11, 12, 45]. This chapter is devoted to this specific problem, looking for a
generalization of the method proposed. We start reviewing the algebraic treatment used in
Chap. 2, analyzing the consequences of parameter choices different from the QNSB ones.
5.1. The algebraic form of the general Morse Hamiltonian

We start deriving the algebraic form of the Morse Hamiltonian in more general conditions.
In Section 2.5.4 we showed that the Morse Hamiltonian can be cast in the simple algebraic
form (87) in terms of the operators (85). We noticed already that the simple factorized form
depends on the particular choice q = s of the SUSY parameter. We derive here explicitly the
algebraic form of the Hamiltonian without imposing special value of the q parameter. For the
moment we will neglect the fact that the s parameter is implicitly involved in the definitions
(85) through the definition (27) of y. This will be dealt with in detail later.
In the following, a more useful expression replaces Eq. (91), (92):

ˆ †
2q I − Â (q) + Â(q)
(129) e−α(x̂−x0 ) = ,
(2s + 1)
~α h †
i
(130) p̂x = Â(q) − Â (q) .
2i
Here we considered for generality the q parameter in the family of operators (85) distinguished
from the s parameter implicitly involved in ŷ, but for simplicity we chose the same value for all
the operators, i.e. we restrict ourselves to the use of one single family of operators. We write
57
58 5. UNLINKING OF BASIS AND POTENTIAL PARAMETERS
2
p̂
explicitly the kinetic operator K̂ = 2m . From Eq. (130) we have
(131)
~2 α 2 ~2 α 2 2
K̂ = − [Â(q) − Â† (q)][Â(q) − Â† (q)] = − [Â (q) + Â†2 (q) − Â(q)Â† (q) − Â† (q)Â(q)] .
8m 8m
Recalling the commutation relations (86), K̂ reduces to
~2 α 2 2
(132) K̂ = − [Â (q) + Â†2 (q) − 2q Iˆ + Â(q) + Â† (q) − 2Â† (q)Â(q)] .
8m
To evaluate the potential-energy operator we start from:
2q Iˆ − [Â† (q) + Â(q)]
(133) e−α(x̂−x0 ) = ,
(2s + 1)
so that
{2q Iˆ − [Â† (q) + Â(q)]}{2q Iˆ − [Â† (q) + Â(q)]}
(134) e−2α(x̂−x0 ) =
(2s + 1)2
1
= {4q 2 Iˆ − 4q[Â† (q) + Â(q)] + Â2 (q) + Â†2 (q) + Â(q)Â† (q) + Â† (q)Â(q)}
(2s + 1)2
1
= 2
{4q 2 Iˆ − (4q + 1)[Â† (q) + Â(q)] + 2q Iˆ + 2Â† (q)Â(q) + Â2 (q) + Â†2 (q)} .
(2s + 1)
So we can write the complete Morse potential contribution V̂ as
1
V̂M (x) = V0 {4q 2 Iˆ − (4q + 1)[Â† (q) + Â(q)] + 2q Iˆ + 2Â† (q)Â(q)
(2s + 1)2
2
+ Â2 (q) + Â†2 (q) − {2q Iˆ − [Â† (q) + Â(q)]}
(2s + 1)
V0 2
(135) = 2
{4q Iˆ − (4q + 1)[Â† (q) + Â(q)] + 2q Iˆ + 2Â† (q)Â(q)
(2s + 1)
+ Â2 (q) + Â†2 (q) − (8sq + 4q)Iˆ + 2(2s + 1)[Â† (q) + Â(q)]
V0
= 2
{(4q 2 + 2q − 8sq − 4q)Iˆ + (4s − 4q + 1)[Â† (q) + Â(q)]
(2s + 1)
+ 2Â† (q)Â(q) + Â2 (q) + Â†2 (q)} .
The complete Hamiltonian ĤM can thus be written as
~2 α 2 2
ĤM = − [Â (q) + Â†2 (q) − 2q Iˆ + Â(q) + Â† (q) − 2Â† (q)Â(q)] +
8m
V0
(136) {(4q 2 + 2q − 8sq − 4q)Iˆ + (4s − 4q + 1)[Â† (q) + Â(q)]
(2s + 1)2
+2Â† (q)Â(q) + Â2 (q) + Â†2 (q)} ,
and finally

V0 ~2 α 2
ĤM = − [Â2 (q) + Â†2 (q)]
(2s + 1)2 8m

V0 ~2 α 2
(137) + 2
(4s − 4q + 1) − [Â† (q) + Â(q)]
(2s + 1) 8m
2 2

2V0 ~α † 2V0 ~2 α 2 ˆ
+ + Â (q)Â(q) + q (2q − 1 − 4s) + I.
(2s + 1)2 4m (2s + 1)2 4m
5.2. HAMILTONIAN AND BASIS PARAMETERS 59
Equation (137) demonstrates that in general, the representation of the Morse Hamiltonian in
a basis of the form
s
αn! y
(138) φσn (y) = y σ e− 2 Ln2σ−1 (y), n = 0, 1, 2, . . .
Γ(2σ + n)
is 5-band diagonal. The expression (137) holds for any choice of the parameters s and q, which
need not be given any specific physical meaning.
The term proportional to [Â2 (q) + Â†2 (q)] is eliminated from ĤM imposing
V0 ~2 α 2
(139) ≡ ,
(2s + 1)2 8m
which amounts to choosing
√
2mV0 1
(140) s= − .
~α 2
In other words, the choice (140) of the parameter s of Eq. (24) makes the Morse Hamiltonian
tridiagonal on the corresponding QNSB-like basis. This holds irrespective of q. Under the
condition (139), the Morse Hamiltonian becomes:
4V0 n † 2 o
(141) ĤM = [ Â (q) + Â(q)] (s − q) + Â †
(q) Â(q) + (q − 2sq) Iˆ .
(2s + 1)2
From Eq. (141), it is clear that by setting
(142) q = s,
the operator form of the Hamiltonian further simplifies, and the Hamiltonian factorizes as:
V0 ˆ,
(143) ĤM = 4 2
[Â† (q)Â(q) − s2 I]
(2s + 1)
which is precisely the representation of Eq. (87).
The previous considerations show that the algebraic form of the Hamiltonian itself is affected
by the choice of the various parameters: the QNSB parameters of Sect. 2.5.4 lead to the
simplest operator form. In view of subsequent applications to one-dimensional problem and for
optimization purposes, a better comprehension of the roles and consequences of the choices of
the parameters is surely useful.
5.2. Hamiltonian and basis parameters

The Hamiltonian and the basis are in principle unrelated concepts. Even within the class of
basis states (138), it is possible to employ different choices of the parameters present in Hamil-
tonian and in the QNSB(-like) basis employed. Of course, this will give a matrix representation
different from the QNSB one; it will induce changes in the algebraic form of the Hamiltonian
and in the basis set. The most general situation occurs when the parameters related to the
Hamiltonian are all different from those present in the basis definition. In the following, we add
subscript P and B to label the parameters in the “potential”, i.e. in ĤM , and in the “basis”
respectively. We rewrite the Hamiltonian (87) as
~2 αP2 †
(144) ĤM = [ÂP (sP )ÂP (sP ) − s2P ] ,
2µ
where sP is the value of s given by Eq. (24) and the choice (142) is implied. Likewise, we
consider the basis
s
αB n! yB
(145) φσn (yB ) = y σ e− 2 Ln2σ−1 (yB ), n = 0, 1, 2, . . . ,
Γ(2σ + n)
where yB = (2sB + 1)e−αB (x̂−x0B ) , and sB is fixed arbitrarily. The operators of Eq. (85) can be
written explicitly in terms of either parameters as
ŷB,P i
(146) ÂB,P (q) = q Iˆ − + p̂x ,
2 ~αB,P
yB,P
ˆ i
Â†B,P (q) = q Iˆ − − p̂x ,
2 ~αB,P
where explicitly
(147) ŷB,P = (2sB,P + 1)e−αB,P (x̂−x0B,P ) .
With the issue of efficiency in mind, a brief analysis of the role of the different parameters,
helps in avoiding unfruitful optimization strategies:
• the α parameter. Consider the recursive relations from Ref. [14]
α αP
− αP αB (x̂−x0 ) − α (x̂−x )
(148) hφσm |e B |φσn+1 i = [Cm hφσm−1 |e αB B 0
|φσn i +
αP α
− P α (x̂−x0 ) σ
(n − m − )hφσm |e αB B |φn i]/Cn+1 ,
αB
αP
σ − αB αB (x̂−x0 ) σ Γ(2σ + ααBP )
(149) hφ0 |e |φ0 i = αP ,
(2s + 1) αB Γ(2σ)
where the Cn are defined in Eq. (48). Equation (148) implies that if αP 6= kαB , for
integer k, all matrix elements of a power of e−αP (x̂−x0P ) are non-vanishing. Moreover,
it is not trivial to obtain such matrix elements, because the recursive relations are less
practical than our algebraic rules. These observations suggest not to allow α to vary,
because this would imply a very high computational cost. Thus, in the following, we
impose
(150) αB ≡ α P .
• the x0 parameter. A shift in the x0 parameter x0P → x0B only produces a trivial
multiplication of all matrix elements of e−α(x̂−x0 ) by a constant factor:
(151) hφσm |e−α(x̂−x0P ) |φσn i = e−α(x0B −x0P ) hφσm |e−α(x̂−x0B ) |φσn i .
The momentum operator is invariant with respect to a shift of the corresponding
variables, and therefore its matrix elements are the same as before.
• the s parameter. The role of this variable is more intricate. We check if it is possible
(†)
to obtain algebraically the matrix elements of an operator ÂP (q) in a basis with
sB 6= sP . From Eq. (146), evaluated once for B and once for P , we write
q + q0 ˆ
(152) e−α(x̂−x0 ) = I − [ÂP (q) + Â†P (q 0 )] ,
2sP + 1
q + q0 ˆ
(153) e−α(x̂−x0 ) = I − [ÂB (q) + Â†B (q 0 )] ,
2sB + 1
5.3. DIFFERENT sB AND sP 61
2sP + 1 −α(x−x0 ) i
Â†P (q) = q Iˆ − e − p̂x
2 ~α
ˆ 1 2sP + 1 0 00 0 † 00 ÂB (q 000 ) − Â†B (q 000 )
(154) = qI − {(q + q ) − [ÂB (q ) + ÂB (q )]} − ,
2 2sB + 1 2
and
2sP + 1 −α(x−x0 ) i
ÂP (q) = q Iˆ − e − p̂x
2 ~α
ˆ 1 2sP + 1 0 00 0 † 00 ÂB (q 000 ) + Â†B (q 000 )
(155) = qI − {(q + q ) − [ÂB (q ) + ÂB (q )]} − .
2 2sB + 1 2
These lead to

† 1 2sP + 1 0 00 ˆ 1 2sP + 1 0 † 00 ÂB (q 000 ) + Â†B (q 000 )
(156) ÂP (q) = q − (q + q ) I + [ÂB (q ) + ÂB (q )] − ,
2 2sB + 1 2 2sB + 1 2

1 2sP + 1 0 00 ˆ 1 2sP + 1 0 † 00 ÂB (q 000 ) + Â†B (q 000 )
(157) ÂP (q) = q − (q + q ) I + [ÂB (q ) + ÂB (q )] + .
2 2sB + 1 2 2sB + 1 2
The q 0 , q 00 and q 000 parameter are in fact free parameters, according to the definition
(85). For simplicity, we choose q = q 0 = q 00 = q 000 ≡ q̃ and then, by defining
2sP + 1
(158) h̃P,B := ,
2sB + 1
we have
Â†B (q̃) ÂB (q̃)
(159) Â†P (q̃) = q̃[1 − h̃P,B ]Iˆ + [1 + h̃P,B ] − [1 − h̃P,B ] ,
2 2
Â† (q̃) ÂB (q̃)
(160) ÂP (q̃) = q̃[1 − h̃P,B ]Iˆ − B [1 − h̃P,B ] + [1 + h̃P,B ] .
2 2
In the case sP = sB , one has h̃P,B = 1, and the identities
(161) Â†P (q̃) = Â†B (q̃) ,
(162) ÂP (q̃) = ÂB (q̃) ,
are recovered. Equations (159) and (160) permit the extraction of the matrix elements
of ĤM and the corrections in Vd in a basis parameterized by different s, by means of
algebraic manipulations. The sparseness of the representation of Ĥ remains granted
even for sB 6= sP .
5.3. Different sB and sP

We have shown that Morse Hamiltonian has the simplest form (143) when written in terms
of the operators (146) if q and sP are those of QNSB. If a different basis is chosen, it is useful
to write the relevant operators for the Hamiltonian in terms of operators whose action on the
basis states is particularly simple, even though this leads to a more complicated Hamiltonian.
We always assume a complete basis of the general form (145): we decide to assume x0B = x0P ,
i.e. refer all the operators to the minimum of the Morse potential, and for simplicity, we also
impose αB = αP , to keep the Hamiltonian matrix sparse. We take instead full advantage of
the freedom of choice of sB , σB .
The basis (145) of states |φσn i determines the operators Â(q) and Â† (q) whose action on
the chosen basis is the simplest. Our task is to write the Morse Hamiltonian and all relevant
operators in term of those operators. The action of the operators Â and Â† is expressed by the
equalities
(163) Â(σ + n)|φσn i = Cn |φσn−1 i ,
Â† (σ + n)|φσn i = Cn+1 |φσn+1 i ,
as soon as the same parameters σB , αB , sB appear in both members. We convert the Morse
Hamiltonian (144) employing Eqs. (158), (159) and (160). Let us introduce the notation
(164) u± = 1 ± h̃P,B .
With this choice Eqs. (159), (160) become
Â†B (q̃) ÂB (q̃)
(165) Â†P (q̃) = q̃u− Iˆ + u+ − u− ,
2 2
Â† (q̃) ÂB (q̃)
(166) ÂP (q̃) = q̃u− Iˆ − B u− + u+ .
2 2
Substituting in Eq. (144), we obtain
~2 α 2 †
ĤM = [Â (sP )ÂP (sP ) − s2P ]
2m P
Â† (sP ) ÂB (sP ) Â† (sP ) ÂB (sP )
(167) = [sP u− Iˆ + B u+ − u− ][sP u− Iˆ − B u− + u+ ] − s2P
2 2 2 2
~2 α 2 ˆ 2 2 sP s u
P + s P u− u− †
= I[u− (sP + ) − s2P ] + u− − − [ÂB (sP ) + ÂB (sP )]
2m 2 2 2 4
† u2+ + u2− u+ u−
+ ÂB (sP )ÂB (sP ) + [Â†2 2
B (sP ) + ÂB (sP )] .
4 4
The matrix elements necessary to compute the Hamiltonian matrix in the basis (145) with
general values of σB and sB are then:
(168) hm|ÂB (sP ) + Â†B (sP )|ni = Cn+1 δm,n+1 + 2(sP − σ − n)δm,n + Cn δm,n−1 ,
(169) hm|Â†B (sP )ÂB (sP )|ni = Cn+1 (sP − σ − n)δm,n+1 + [Cn2 + (sP − σ − n)2 ]δm,n
+ cn (sP − σ − n + 1)δm,n−1 ,
(170) hm|Â2B (sP ) + Â†2

B (sP )|ni = Cn+2 Cn+1 δm,n+2 + Cn+1 (2sP − 2σ − 2n − 1)δm,n+1
+ 2(sP − σ − n)2 δm,n
+ cn (2sP − 2σ − 2n + 1)δm,n−1 + Cn Cn−1 δm,n−2 .
Using relations (94), (168), (169), and (170), the matrix elements of any distorted Morse
Hamiltonian of the form of Eq. (33) can be computed algebraically, using the same formal
expressions as for QNSB.
5.4. Bases and efficiency

We can apply the formalism of previous section to verify the efficiency of families of wave-
functions (138) with different parameter choices.
The QNSB parameters have a few advantages over different choices:
(1) s definition (24) and sB = sP assumption makes the Morse Hamiltonian tridiagonal;
(2) the choice q = s in Eq. (85) factorizes the Morse Hamiltonian;
5.4. BASES AND EFFICIENCY 63
3 3
2 (a) 2 (b)
1 1
σ σ
φ 4(x) φ 4(x)
0 0
-1 -1
-2 -2
0 2 4 6 0 2 4 6
αx αx
3 3
2 (c) 2 (d)
1 1
σ σ
φ 4(x) φ 4(x)
0 0
-1 -1
-2 -2
0 2 4 6 0 2 4 6
αx αx
Figure 5.1. Variation of the 5th QNSB wavefunction for a 30% increase in
the parameters s (a), σ (b), x0 (c) and α (d) (solid line). The QNSB original
wavefunction is shown for comparison (dashed line), and corresponds to s = 8.34,
σ = 0.34, x0 = 1, α = 4.
(3) the choice σ = s − [s] forces a natural decomposition of the Hilbert space that allows
an isomorphism of the Morse basis and the subspace generated by the first [s]+1 states
in the QNSB [14].
These are indeed great formal results, but in the practice of computational spectroscopy,
they are perhaps not so important. When we apply this method to a realistic potential of the
type (34), some of the previous features are no more real advantages:
• the QNSB does not provide an exact solution, and requires diagonalization anyway;
• the matrix representation of the potential correction Vd is at least 5-band diagonal;
• if the potential is substantially distorted from the Morse one, the actual eigenfunctions
may be represented poorly by the Morse bound states, or equivalently by their QNSB
counterparts: it may then be preferable to allow for substantial mixing of the Morse
bound and unbound Hilbert subspaces, already in the basis;
• in subsequent applications to multi-dimensional systems, the choice of QNSB for indi-
vidual anharmonic oscillators may not be particularly efficient for the coupled problem,
as significantly as more basis states are often required to describe an oscillator with
few bound states.
With these considerations in mind, we check now for advantages in choosing parameters different
from the QNSB ones. We have no a priori recipe to choose the alternative parameters; to gain
some insight, a brief analysis of Morse basis and QNSB-like parameter dependence can be
useful.
Figure 5.1 shows the behavior of a wavefunction of functional form (43) after variation of
the parameters involved. Observe that:
• The dependence on the s parameter (Fig. 5.1a) is weak. By increasing s, the eigen-
function shifts almost rigidly towards the outer region.
• Also the dependence on σ (Fig. 5.1b) is weak, but σ can vary in a wide range: increas-
ing σ the wavefunction deforms and shrinks, concentrating towards the region of the
minimum, and vanishing more rapidly for large x values.
• The variation of the x0 (Fig. 5.1c) parameter causes the expected translation.
• Increasing the α parameter (Fig. 5.1d) causes the wavefunction to shrink. It concen-
trates near the minimum, and vanishes more rapidly for large x values. This depen-
dency is strong.
From the considerations above we can make some guesses. For example, inspection of
Eq. (103) indicates that the smaller σ is, the more extended the wavefunctions are. This
suggests that a small σ value would make the basis wavefunctions spread away from the well
region. As the states closest to dissociation are by nature slowly decaying outside the classical
limits, small σ could improve the convergence of high-energy states. To address the problem
of the optimal choice of parameters QNSB-like basis, we want the following constraints to be
satisfied:
(a) the Hamiltonian matrix should remain as sparse as possible. This is best achieved if
αB = αP , and is completely lost if ααBP is not integer.
(b) A variation in s reflects a variation in V0 . The dependence of wavefunction on s is not
strong, but the variation in s does not affect the sparseness of the matrix, thus we can
as well vary it as a free parameter.
(c) The variation in σ strongly affects the shape of the wavefunctions. If the whole eigen-
value spectrum is addressed, including near dissociation states, we make the safe choice
that 0 < σ < 1.
In practice, we refer for example to the pure Morse potential. As a first check, we adopt
parameters from Ref. [9]. In that paper, in the context of multidimensional problems, Tennyson
and Sutcliffe propose the use of a complete basis that formally coincides with that of Eq. (43).
We came across their work only recently, and realized that these authors use the basis of
Eq. (43), recognizing its properties only partly, as an intermediate tool in their computational
approach. Their choice for the parameters is very different from those of the QNSB. Written
in our notation, in the case of Morse potential Tennyson and Sutcliffe choose a large
[2s] + 2
(171) σ≡ .
2
This means that for large x the wavefunctions they employ vanish at least as rapidly as
[2s]+2 [2s]+2
y 2 = [(2s+1)e−α(x−x0 ) ] 2 . Direct application of that choice gives a measure of its accuracy
and efficiency. Figure 5.2 shows the difference between the eigenvalues obtained diagonalizing
the corresponding matrix and the exact value, versus the basis dimension, for a pure Morse
potential (~ = 1, µ = 1, s = 8.34). The convergence of the three highest-energy states requires
hundreds of basis states, while the QNSB choice of the parameter gives the exact result with
only [s] + 1 = 9 basis states.
A spontaneous objection is that the pure Morse potential is a very special case, being
a solvable potential. Moreover, Tennyson and Sutcliffe treat the parameters as variational
quantities, to be adjusted to obtain better convergence properties for the single potential.
Tennyson and Sutcliffe’s choice does not respect condition (a), but they are not interested
to keep matrices sparse, as they employ a discrete-variable representation [18] to treat the
potential. They also do not respect condition (c), as they impose σ = [2sP2]+2 , so that σ is
an integer. According to our previous analysis, this should give wavefunctions concentrated
σ=9
0.1
level 8
- E )/V0
0.01
ex
level 7
alg 0.001
(E
0.0001
level 6
0 200 400 600 800 1000
Ns
Figure 5.2. Convergence of the Morse potential eigenvalues (logarithmic scale)

with respect to QNSB size with the convention of Ref. [9, 11] (~ = 1, µ = 1,
s = 8.34).
1 1 1
QNSB c) QNSB
a) QNSB b)
0.01
- E )/V0
0.01 0.01
- E )/V0
- E )/V0
ex
ex
ex
0.0001 0.0001 0.0001

alg
QNSB+ QNSB+
alg
alg
(E
(E
(E
QNSB+
1e-06 1e-06 1e-06
0 5 10 15 0 5 10 15 0 2 4 6 8 10
n n n
Figure 5.3. Discrepancies (Enalg − Enex )/V0 of the individual eigenvalues n for
the three potentials V (x) = VM (x) + 1.0 V0 [v(x)]4 (a) (QNSB+ with sB = 23.36,
σB = 0.435); V (x) = VM (x) + 1.0 V0 [v(x)]6 (b) (QNSB+ with sB = 29.55, σB =
P
0.360); and V (x) = 2.0 V0 6i=3 [v(x)]i (c) (QNSB+ with sB = 27.46, σB = 0.430),
for all the bound states, using QNSB and a QSNB+.
in the well region, and indeed to improve the convergence the authors claim that empirically
...The optimal value of re (≡ x0 ) is usually larger than the value given by the minimum of the
potential;... the values for re and ωe (∝ α) are often quite strongly coupled and need to be varied
together. In fact we understand that the modifications in x0 and αB envisaged by Tennyson
and Sutcliffe attempt to balance a very poor correlation imposed between σ and s.
Here, rather than making a priori choices of the parameters s and σ, we try to tune them
to obtain best convergence of the levels for a fixed number Ns of basis states, or to obtain
convergence of all bound levels with as small a basis size as possible. We apply this optimized
QNSB, which we refer to as QNSB+, approach to the cases discussed in Section 2.5.7 above.
0.0001 0.0001
4
8e-05 8e-05 V(x) = VM(x) +0.5 V0 [v(x)]
4 Ns = 30, sB = 21.33
∆ V(x) = VM(x) +0.5 V0 [v(x)] ∆
6e-05 Ns = 30, σB = 0.185 6e-05
4e-05 4e-05
2e-05 2e-05
a) b)
0 0
16 18 20 22 24 0.15 0.18 0.21
sB σB
Figure 5.4. sB dependence of ∆ (173) for best σB a), and σB dependence of

for best sB b), for V (x) = VM (x) + 0.5 V0 [v(x)]4 , Ns = 30.
We thus compute the spectra for the three potentials

(172) V (x) = VM (x) + 1.0 V0 [v(x)]4 ,
V (x) = VM (x) + 1.0 V0 [v(x)]6 ,
6
X
V (x) = 2.0 V0 [v(x)]i .
i=3
We consider the same x0 and α parameter, and treat sB and σB as free parameters. Convergence
is measured by the total mean square discrepancy of the bound eigenvalues
Nb −1
1 X
(173) ∆= (E alg − Eiex )2
Nb i=0 i
from the exact ones (with Nb = number of bound states). We minimize the expression (173)
with respect to the two parameters sB and σB , for a given number Ns of basis states, the
same as in Sect. 2.5.7 (i.e. Ns = 30 for the potentials of Fig. 5.3a) and c), and Ns = 40 for
Fig. 5.3b)), which produced the spectra of Tables 2.13, 2.14 and 2.15. Figure 5.3 shows the
Alg Ex
values of (EN s
− EN s
)/V0 for the three potentials, for QNSB and an optimized QNSB+ basis.
The QNSB, for all the three potential has sP = 8.33(8) and σP = 0.338, while the optimized
basis parameters sB and σB are reported in the caption of Fig. 5.3 for each potential.
As can easily be checked, choices of the parameters different from the QNSB ones allow
a substantial improvement of the accuracy of the results, with the same computational cost.
In other words, the convergence speed of the computation can be drastically improved with
a suitable choice of the parameter sB and σB . Moreover, the sB dependence of the total
discrepancy ∆ is typically that illustrated in Fig. 5.4 a): for σB equal to its “best” value,
approaching the optimal sB value from below ∆ decreases relatively slowly, and for values
greater than the best value, it grows very steeply. The σB dependence of ∆, Eq. (173), is
illustrated in Fig. 5.4 b): for sB set to the best value, there is a very deep minimum at
some 0 < σB < 1. Figure 5.5 shows a 2-dimensional view of ∆ in the neighborhood of the
0.100
0.075 27.36
∆ 0.050
0.025 25.36
0
23.36
PSfrag replacements 0.235
0.435 sB
21.36
σB 0.635
0.835
Figure 5.5. 2-dimensional picture of ∆ near its minimum region for V (x) =
VM (x) + 1.0 V0 [v(x)]4 , Ns = 30. Observe the remarkable asymmetry of the sB
dependence.
Ns sB σB
18 20.82 0.449
22 24.54 0.440
26 24.75 0.428
30 27.46 0.430
15 16.7568 3.178
Table 5.1. sB and σB parameterP6 minimizing ∆, defined in Eq. (173), for the
i
usual potential V (x) = 2.0 V0 i=3 [v(x)] . Varying the number Ns of basis states,
sB vary consistently, while σB is a better defined quantity. In the first four rows
all the 11 bound states of the expanded potential are addressed. In the last row
only the lowest eight bound states are addressed: notice that the σB value is
much larger here.
minimum: notice the steep well along the σ coordinate, and the asymmetric dependence on the
sB coordinate.
These considerations suggests that the relation (96) σB = sB − [sB ] is not satisfied in the
general case. The usefulness of the relation (96) is indeed questionable: it allows to identify the
Morse bound-states-generated subspace with that generated by the first [s(P ) ] + 1 QNSB states.
In contrast, the optimized QNSB+ renounces to a tight relation with Morse basis in favour
of a better resemblance to the basis of eigenstates of the expanded potential; accordingly, the
property implied by Eq. (96) is not especially useful or even meaningful when the potential is
not Morse.
As the accuracy seems much more sensitive to the σB than to the sB parameter, the rela-
tion (96) could be forced, choosing the sB value that gives the highest accuracy among those
satisfying sB − [sB ] = σB , where σB minimizes ∆, defined in Eq. (173). Such sB value would
allow to identify through Eq. (24) a family of Morse potentials bearing to QNSB+ the ordinary
Morse basis-QNSB relations. In a certain way, these could be considered a sort of best Morse
potentials in approximating the target one. The similarity is no more in the shape in coordinate
space, but in the degree of overlapping of bound-states spectra.
0.9 QNSB+
QNSB
Projection
0.8
4
0.7 VM(x) + 0.5 [v(x)]
0.6
0 20 40 60 80
Ns
P s −1 σ
Figure 5.6. The projection N 2
j=0 |hφj |ϕk i| of the highest eigenstates on the
subspace spanned by the QNSB basis (blue line, triangles), and on the generalized
QNSB+ (red line, diamonds), for the weight R = 0.5 of the deformation Vd (x) =
[v(x)]4 added to a Morse potential characterized by [s] + 1 = 9 bound states, with
different number Ns included in QNSB (~ = 1, µ = 1, s = 8.34). For each Ns a
different generalized QNSB is chosen. The improvement is substantial.
We also address again the tough case R = 0.5 of Fig. 2.16, where the extremely close vicinity
PNs −1
of the highest bound state to dissociation makes its projection j=0 |hφσj |ϕk i|2 on QNSB very
PNs −1 σ 2
slowly converging with the QNSB size. We compute the projection j=0 |hφj |ϕk i| , this
time using for optimized parameters (sB , σB ), and compare the same different basis sizes Ns
considered in Fig. 2.16. Incidentally, we obtain different sB values but similar σB for different
Ns (Table 5.1). Figure 5.6 illustrates the substantial improvement in the projection: practically
unity projection is already obtained for basis size Ns = 18 only.
Alternatively one can find the best (sB , σB ) values that give the most accurate spectrum
with the smallest number of states. For example, we looked for the minimum number Ns of
basis states sufficient to compute the complete bound state spectrum of the potential V (x) =
VM (x) + 1.0 V0 [v(x)]4 . We find that taking sB = 13.56 and σB = 0.047, Ns = 15 states
are enough to give the exact number of bound states, but with a ∆15 = 0.9%. With only
one additional basis state, the accuracy improves substantially: for sB = 14.47, σB = 0.308
and Ns = 16, the whole spectrum, as shown in Table 5.2, is obtained with an acceptable
∆16 = 0.034%.
As a final observation, we already mentioned the fact that Morse and QNS basis behave
differently if the number Ns of states included in the basis is less than the number of bound
states of the target potential. The Morse basis in general keeps giving good results for low
energy states, and fails badly on the near dissociation ones: only few low-energy energy states
are identified correctly (see Table 4.3). A poorly converged QNSB instead produces a larger
number of bound states, but they are all obtained with rather poor accuracy. Equation (103)
and Figures 2.10 and 2.11 clarify the reason of this behavior. The Morse ground state is
localized in the well region and excited bound states extend to larger x distances. In contrast,
the QNSB 0th wavefunction is rather broad, while the excited ones are less. Truncation of the
state Finite Differences QNSB+ (Ns = 16)

0 -551.3534 -551.3525 (2·10−4 %)
1 -407.9164 -407.9121 (7·10−4 %)
2 -266.8248 -266.8135 (0.0018%)
3 -131.4116 -131.3886 (0.0036%)
4 -3.7564 -3.7137 (0.0068%)
5 114.6639 114.7420 (0.012%)
6 222.7122 222.8449 (0.021%)
7 319.4620 319.6414 (0.029%)
8 404.1247 404.2970 (0.028%)
9 476.0042 476.1627 (0.025%)
10 534.4662 534.6792 (0.034%)
11 578.9155 579.1094 (0.031%)
12 608.7773 608.9047 (0.020%)
13 623.4814 623.6523 (0.027%)
Table 5.2. Potential expansion with Vd (x) = 1.0 V0 [v(x)]4 , (α = 4, x0 = 1, µ =
1, ~ = 1, V0 = 625), so that the dissociation threshold is V0 : comparison of
results of diagonalization of tridiagonal QNSB+ (Ns = 16) and finite differences
Alg Ex
method. In parentheses (EN s
− EN s
)/V0 . ∆16 = 0.034%.
Morse basis thus allows to reproduce better mainly localized states, while highly excited states
are inaccurate; truncation of the QNSB affects the accuracy of all states almost equally.
This can be seen indirectly if we allow the parameter sB and σB to vary. If we try to find
the value for σB that allows a better reproduction of the K lowest-lying energy states, we find
that that optimal σB values lie in the neighborhood of Nb − K, where Nb is the number of
bound states of corrected potential (Table 5.1). As the effect of increasing σB is to produce
more localized wavefunction, this choice improves the accuracy of the eigenvalues obtained. In
such a way, the low-energy states are reproduced fairly well with few basis states even with the
few first eigenstates of a standard QNSB+ basis, but the convergence of the high-energy states
requires a lot more states, as shown in Figure 5.2.
As a final remark, note that above we have derived the optimal parameters sB and σB for the
QNSB+ basis by minimizing ∆, Eq. (173), whose definition requires the knowledge of accurate
eigenvalues Eiex (e.g. obtained by an independent method, or by a fully converged QNSB
calculation). In fact, the previous knowledge of the exact eigenvalues is not necessary: due to
the variational nature of the approach, a more efficient basis produces lower eigenvalues, so the
sB and σB parameters can be obtained as those that produce the lowest eigenvalue spectrum
for the assigned basis size Ns . This approach involves knowing the number of eigenvalues to
consider in the minimization process. To address the entire bound state spectrum, one should
know the exact number Nb of bound states, which can be estimated by means of a sufficiently
large QNSB.
The QNSB+ basis becomes very useful in polyatomic context, where individual local oscil-
lators are first solved exactly (for example on a very wide QNSB), then an optimal QNSB+ is
generated for each of them, thus involving few states only, and finally the coupled molecular
problem is solved in the product basis of the individual local modes.
CHAPTER 6
Interacting oscillators
Application to polyatomic molecules of the method proposed in the present work is not
straightforward. Further developments and refinements would require accounting for different
interactions which presently are not included in any sense exactly in our scheme. Even rotational
effects are not yet included in our treatment, as they are not in every other approach based on
Morse potential wavefunctions we are aware of [26, 27, 28, 29, 30]. On the other hand, all
approximate solutions attempted in those models can be applied here.
Despite still missing some needed extensions, our model has some features that makes it a
perspectively interesting tool for polyatomic molecules:
• includes anharmonicity from the beginning;
• applies to a rich class of one-dimensional potentials;
• is (substantially) free from approximations;
• gives exact results in the limit of infinite basis size;
• the algebraic approach makes it a computationally efficient method.
In a sense this method shares several of the characteristics of simplicity and manageability with
the harmonic-oscillator-based expansion.
There are mainly two different playgrounds where this method could be applied. One is
that of diatomic molecules, that is still an interesting ground by its own. Modern highly
resolved IR and MW spectra provide accuracy in line positions of the order of 10−3 cm−1 ,
10−7 cm−1 respectively, and so are a testing ground for all minor details of theoretical models.
Such a high accuracy requires obviously a very accurate modeling of the potential energy
surface, but also the inclusion of roto-vibrational interactions, and of non-adiabatic effects
[19, 20, 21, 46, 47, 48], and, in turn, these require further theoretical work.
A more interesting and challenging field is that of polyatomic molecules. Currently employed
models in physics and quantum-chemistry allow the description of triatomic [10, 11, 12, 49],
tetratomic molecules, while attempts to deal with great accuracy with systems with more
than four atoms are still not particularly general and reliable: some models applying to specific
molecules have been proposed [50, 51, 52, 53, 54]. Thus the development of more sophisticated
molecular models is still the main route to obtain more accurate and more efficient theoretical
tools. In these molecules the distance between experimental results and theoretical predictions
is far larger than for the simple diatomic molecules, and there is much room for attempts to
reduce it.
One of the main differences with respect to the diatomic case, is that the increasing number
of degrees of freedom involved in polyatomic systems leads to more intricate forms of both
kinetic and potential energy in the quantum mechanical Hamiltonian. This complexity reflects
the possibility of a wide class of interactions among the different degrees of freedom, of kinetic
and potential origin, in principle all important for an accurate description of the molecules.
In particular, the choice of different coordinate systems for the description of the molec-
ular dynamics leads to very different formal expression for the kinetic operator. Generally, a
coordinate system suitable to represent efficiently the potential contribution makes the kinetic
operator form almost unmanageable, and the introduction of simplifying approximation seems
71
72 6. INTERACTING OSCILLATORS
unavoidable. The comparison among different methods becomes sometimes very difficult, as
the effect and even the meaning of some simplifications may be only guessed.
Nonetheless, several models, methods, and techniques have been proposed and employed.
Two main families can be identified. One follows at least in its philosophy the traditional
normal mode approach, considering a zero-order uncoupled harmonic-oscillator Hamiltonian,
with correction of ever increasing complexity, and even with some attempts towards hybrid
approaches, in which most modes are treated harmonically and few others with more sophisti-
cated techniques. To cite only some of these recent developments, we can mention [2, 12, 45].
The basic assumption is that the most important contributions to molecular spectra are due
to the interbond interactions among different modes, rather than to the peculiar nature of the
individual internal degrees of freedom.
On the other hand, present-day spectroscopy reveals that in the so called chemically sig-
nificant energy region many molecules show relatively simple regularities, in contrast with the
expected rich and intricate structure typical of strongly interacting closely spaced states. These
regularities are best described adopting the concept of local mode vibration: especially high
energy excitations of many polyatomic molecules often result in localized excitation of a single
(usually) stretching oscillator, whose precise features become the essential ingredient for an
accurate description. The localized excitation and the related energy transfer among different
bonds can also be relevant for dissociation phenomena [14, 55]. In this condition, a more
natural description is obtained introducing internal coordinates (rather than normal modes).
The kinetic operator Hamiltonian expressed in terms of internal coordinate is much more
complicated than for normal modes: the complete expression is difficult to obtain [45], or
difficult to manage in the selected basis, thus some terms are neglected from the beginning.
Most of the approaches proposed employ an approximate expression [9, 10, 11]. Another
problem is that no simple recipe allows to separate the internal degrees of freedom from the
collective pure rotation and translation in an arbitrary coordinate systems. As a consequence,
spurious modes can appear, though some attempts to avoid them has been pursued [49].
Our method is probably more suitable to be employed in the local-modes general scheme,
as it provides an accurate and relatively computationally light description of anharmonic in-
teractions, though it does not solve the two main difficulties mentioned above.
We now investigate if it is possible to integrate the QNSB method in existing approaches:
in principle it should be possible for every method already employing a Morse basis expansion
(of the basis or of the potential). In the present preliminary analysis we focus on probably the
simplest model available.
6.1. Multidimensional potential Morse expansion

The choice of internal coordinates is often inspired by the necessity to express the potential
contribution in an accurate though simple form. Among the different coordinates proposed,
the Morse coordinate is used in a variety of different models for polyatomics. For example,
the DVR3D program for triatomic molecules [11] uses such an expansion, but it employs a
“Discrete Variable Representation” for the potential [18]. This makes the potential operator
diagonal, whichever the basis chosen is, so the coordinate choice is only partially exploited.
Several other techniques developed for treating polyatomic molecules use the Morse basis as an
ingredient for treating the vibrational problem. Beside the natural role in the description of the
vibrational contribution to the potential, the need to simplify the whole Hamiltonian induced
many authors to renounce to the exactness of the kinetic part of the Hamiltonian and to adopt
an approximate expression in which also an expansion of non purely potential term in Morse
coordinate y (27) is widely used.
6.2. THE HCAO MODEL FOR TRIATOMIC MOLECULES 73
In Ref. [49] the author writes a Hamiltonian for the H2 O molecule as a expansion up to
th
4 order in the three internal coordinates yi representing both the stretching and the bending
motion. The approach implied is algebraic in nature, and a sophisticated analysis is carried
out to simplify the problem without losing accuracy.
In Ref. [10], the author adopts a potential represented as an expansion in terms of two
Morse coordinates y1,2 for the stretching modes up to 4th -order terms and a bond coordinate
ρ, assuming that this potential expansion is flexible enough to approximate well real triatomic
potential. Then the kinetic energy dependence on stretching coordinates is also expanded in
power series of the Morse coordinate, thus leading to an approximate kinetic-energy operator.
The previous examples show that Morse basis and Morse coordinate expansion are of rel-
atively common use and interest in the context of polyatomic molecules. The possibility of
describing anharmonicity efficiently and the availability of exact analytic expression for the
matrix elements, encouraged the use of this basis even though the Morse basis does not allow
all simplifications of the harmonic oscillator. In some of the mentioned (and in other) methods,
the simplifications brought by QNSB approach can be easily incorporated.
Two main differences emerge with respect to the simple diatomic case:
• in the polyatomic case several vibrational and rotational degrees of freedom interact.
The Morse potential is a good starting point to approximate a typical stretching po-
tential, but its use for bending degrees of freedom is questionable, although it has
sometimes been used [49]. Moreover, coordinates dependency appears also in the
kinetic energy term, and this can hardly be treated exactly;
• the assessment of the correct behavior of Morse expansion in the whole range of the
stretching coordinates for a multidimensional potential expansion is not trivial.
In the following Section, we apply the QNSB formalism to probably the simplest model ad-
dressing polyatomic anharmonic vibrations: the Harmonically Coupled Anharmonic Oscillator
(HCAO) model. This simple model is mainly used to investigate semi-quantitative features of
intermode vibrational energy redistribution, and energy level patterns of polyatomic molecules.
We do not expect the application of our method to improve the accuracy substantially. Nonethe-
less, with the purpose to check if the advantages of our method can be at least partly transferred
in the polyatomic case, the simple HCAO model can be a suitable starting point. Although of
little practical relevance, an interesting advantage of the QNSB for the HCAO is that it allows
one to solve the model exactly, leaving out the simplifying assumptions to which its name is
due. Of course, this is yet a very preliminary step, a long way to a quantitative description of
realistic polyatomic spectra.
6.2. The HCAO model for triatomic molecules

The HCAO model [56] was one of the first attempts to go beyond the usual harmonic-
oscillator-based theory for molecular vibrations. We go into some details to explain the general
features, in a special case.
Traditionally, for a triatomic molecule of the form H2 X, the HCAO implies diagonalizing
N + 1 dimensional matrix blocks in the normal mode basis |v1 , v2 , v3 i, where the v1,3 are
stretching and v2 is a bending mode. The diagonal elements are given by the Dunham terms
3 3 X
3
X 1 X 1 1
(174) hv1 , v2 , v3 |Ĥvib |v1 , v2 , v3 i = ωi vi + + xij vi + vj + ,
i=1
2 i=1 j=1
2 2
while the off-diagonal matrix elements are of the Darling-Dennison type:

1
(175) hv1 + 2, v2b , v3 − 2|Ĥvib |v1 , v2 , v3 i = K11,33 [(v1 + 1)(v1 + 2)v3 (v3 − 1)]1/2 ,
4
1
hv1 − 2, v2b , v3 + 2|Ĥvib |v1 , v2 , v3 i = K11,33 [v1 (v1 − 1)(v3 + 1)(v3 + 2)]1/2 .
4
In these expansions, Ĥvib is approximated by products of harmonic-oscillator Hamiltonians.
Each matrix block has one common v2 value and a common value of N = v1 + v3 . The normal-
mode-based approach that leads to Eq. (174) treats the molecule initially as three uncoupled
harmonic oscillators. The zero-order model includes a partial treatment of the interactions
between different vibrations described in terms of internal coordinates, as these interactions
are included in the normal coordinates definition, at the harmonic level. All anharmonicities
of the molecular potential are treated as perturbations, assuming that they only cause slight
shifts of the zero-order harmonic energies. The model is expected to produce accurate results in
molecular systems where this assumption is reasonable, which means only for the states much
lower than the dissociation threshold.
In most molecules (even triatomic) such as H2 X or other hydrides anharmonicity plays a
main role. This is particularly true for highly excited stretching states, when the region of
potential surface of a single stretching oscillator is explored. In such cases, anharmonicity
contributes heavily to the total energy. On the other hand, the interaction between different
internal coordinates may be not so important. A simple model that tries to address this
situation is the HCAO [56]. This model incorporates anharmonicity from the beginning, relying
on Morse rather than harmonic oscillators. The zero-order Hamiltonian is
X ~2 ∂ 2
0 2
(176) ĤHCAO = − + V0 [v(xi )] ,
i=1,3
2µ ∂x2i
where [v(xi )]2 is an obvious 2D generalization of Eq. (23), and µ = mmHH+m mX

X
is the reduced mass
of the individual stretching dimer.
It thus consists in a pair of equal independent Morse oscillators, labeled 1 and 3, in the
displacement coordinate xi , representing the two stretching modes. The bending mode is
assumed to be decoupled, and so it is treated separately. Bond anharmonicity is thus taken
into account through the pure Morse potential. The interaction between the two oscillators is
successively taken into account as a secondary contribution, modelled as

cos θ ∂ ∂
(177) Ĥ1,3 = C1,3 v(x1 )v(x3 ) + −i~ −i~ ,
mX ∂x1 ∂x2
where C1,3 is a potential energy parameter and θ is the equilibrium bond angle for the molecule.
Thus coupling between the two oscillators is realized by a potential interaction plus dominant
kinetic interaction. The complete Hamiltonian is
0
(178) ĤHCAO = ĤHCAO + Ĥ1,3 .
This simple Hamiltonian can be diagonalized in its natural basis, the product of Morse oscillator
eigenfunctions, Eq. (26)
(179) hx1 |hx3 |mi|ni ≡ Ψm (x1 )Ψn (x3 ) .
We summarize here the main features:
0 0
• ĤHCAO is diagonal, and its matrix elements are ĤHCAO m,n = ωM [m + n + 1] −
2 2 2
ωM xM [(m + 12 )2 + (n + 21 )2 ], where ωM = ~α 2Vµ0 , ωM xM = − ~2µα , and µ is the
reduced mass of the dimer.
6.2. THE HCAO MODEL FOR TRIATOMIC MOLECULES 75
• Ĥ1,3 matrix elements can be reduced to products of single-oscillator matrix elements:
(180) hΨm0 Ψn0 |Ĥ1,3 |Ψm Ψn i = C1,3 hΨm0 |v(x1 )|Ψm ihΨn0 |v(x2 )|Ψn i +
~2 ∂ ∂
− cos θ hΨm0 | |Ψm ihΨn0 | |Ψn i .
mX ∂x1 ∂x2
All these 1-dimensional matrix elements can be computed using for example Eq. (36). Beside
these, we need matrix elements of derivation operators, and these are available in the literature
(see, for example, Ref. [6, 31]):
∂
(181) hm| |mi = 0 ,
∂x1
∂ ∂ 1
hm0 | |mi = −hm| |m0 i = (m0 |m), m0 > m ,
∂x1 ∂x1 2
where
"
2s+1
2s+1
2s+1
# 12
0 − 2m0 − 1 − 2m − 1 m0 ! Γ − m0
(182) (m0 |m) = (−1)m −m α 2 2 2
,
m! Γ 2s+1
2
− m
and s is defined in Eq. (24). As already discussed in Sect. 2.3, all off-diagonal matrix elements
hm0 |v(xi )|mi and hm0 | ∂x∂ i |mi are non-vanishing. The corresponding normal-mode expression
for one-dimensional harmonic oscillator reads:
∂
(183) hv 0 |Q|vi = hv 0 | |vi = 0 if |v 0 − v| 6= 1 .
∂Q
This major difference allows substantial simplifications: the restrictive harmonic-oscillator se-
lection rules permit great simplifications that are very useful to develop molecular potential
models.
In order to obtain more manageable matrices it is standard to introduce three main approx-
imations in the HCAO model, in addition to the simplifications involved in the Hamiltonian
definition (178):
a) As the energy separation between diagonal matrix elements of basis states differing in
N = m + n is expected larger than those involving off-diagonal terms, the condition
(184) hΨm0 Ψn0 |Ĥ1,3 |Ψm Ψn i = 0 if m0 + n0 6= m + n
is imposed. This produces a substantial basis reduction, analogous to the one produced
by the corresponding harmonic-oscillator “selection rules” (174), (175).
b) The condition
(185) (m0 |m) = 0 if m0 6= m + 1
enforces a relation analogous to Eq. (183) for each harmonic oscillator. This is hardly
justified, and often a poor approximation.
c) For m0 = m + 1, assuming that 2s+14
m, it is assumed that:
r
2s + 1 √
(186) (m + 1|m) = −α m+1.
2
2s+1
This means that the harmonic limit is used, which works well for m 4
, but may
create troubles near dissociation.
Rather than in the Morse eigenfunction basis we can expand the solution in terms of QNSB
(43):
(187) hx1 hx3 |mi|ni ≡ φσm (x1 )φσn (x2 ) .
0
Exact matrix elements of ĤHCAO have been already obtained: the 1D matrix is tridiagonal. 1D
matrix elements of [v(xj )] can be obtained using Eq. (98). Exact matrix elements of ∂x∂ i , and
i
all powers of this operator, can be obtained using Eq. (92), i.e.:
∂ α
(188) hm| |ni = [Cn+1 δm,n+1 − Cn δm,n−1 ] .
∂xi 2
It is apparent from Eqs. (98) and (188) that those matrix elements are non-zero only if
|n − m| ≤ 2i + 1. Accordingly, the QNSB yields an exact matrix representation for the complete
Hamiltonian (178) with less than a few per cent non-vanishing matrix elements. Sparseness is
obtained exactly, not at the price of some approximation, as is customarily done in the Morse
basis. A minor drawback is that the application of the HCAO reduction rules are slightly less
straightforward, as the decomposition of the exact eigenfunctions requires the calculation of
the eigenstates of the two individual oscillators. This is even more relevant if more than [s] + 1
QNSB states are employed.
The QNSB represents an even greater improvement when the individual potentials are
distorted Morse oscillators: in such cases, the Hamiltonian (178) in Morse basis produces a
non-sparse matrix, even without considering the intermode interactions. QNSB matrix instead
still yields a sparse matrix. The number of off-diagonal non-vanishing terms increases with the
degree of the potential expansion (34), but in all practical cases a substantial fraction of the
matrix elements vanishes.
6.3. Comparisons
0
We focus on the model Hamiltonian of HCAO Ĥ = ĤHCAO + Ĥ1,3 , without imposing any
simplification. The inclusion of the interbond coupling can induce a lowering of the dissociation
threshold, due to the lowering of the preferential dissociation
channels. In the case at hand,
(0) 1 C1,3 (0)
the two channels corresponding to x1 = x1 − α log 2V0 + 1 , x3 → +∞, and x3 = x3 −

1 C1,3
α
log 2V0
+ 1 , x3 → +∞ deform and lower in energy as C1,3 increases. The energy lowering
of the “valley” with respect to the non-interacting case is explicitly
2
C1,3
(189) Egain = .
4V0
In addition, this dissociation path consists in a sharp valley, which implies a zero-point local-
ization energy in the direction transverse to the dissociation path which in practice opposes
dissociation. This zero-point energy in the harmonic-approximation is given by
2 1/2
Transv ~α C1,3
(190) E0 = 1 2C1,3 + + 2V0 .
2µ 2 2V0
These two contributions combine to determine the dissociation threshold for the coupled po-
tential:
(191) Ediss = −V0 − Egain + E0Transv
on a scale where the bottom of the well is −2 V0 .
For a first test, we arbitrary choose Morse parameters V0 = 625, α = 3.5, x0 = 1, so that
in units where ~ = 1, MH = 1, we have s = 9.300, so that each oscillator is associated to 10
6.3. COMPARISONS 77
0.08
Non-vanishing ratio
0.06
0.04
0.02
0
10 20 30 40 50
Ns
Figure 6.1. Sparseness of the HCAO Hamiltonian (178) matrix representation:

the ratio of non-vanishing to total matrix elements, as a function of the single
oscillator basis size Ns (the actual matrix side is Ns2 ).
bound states. We compute the eigenvalue spectrum using ordinary Morse basis, and QNSB of
Sect. 2.5.5. We use coupling constants C1,3 = 50, cos α = 0.05, and mX = 16, which gives matrix
elements of the order 10% of the separation between Morse low-energy states. Thus interaction
effects should be relatively small with respect to the independent oscillators features. Table 6.1
reports the results obtained using Morse basis. The effect of coupling is to split the otherwise
degenerate states. As expected, the net effect is small for the chosen coupling parameters. The
use of a Ns = 10 QNSB produces exactly the same results as Morse, as it must. Increasing Ns to
20 or larger improves the convergence, but does not change the spectrum much. It is important
to remark that Ĥ on the Ns = 10 QNSB has only 784 non-vanishing matrix elements, instead
of the 10000 non-zero matrix elements in the Morse-basis representation. The Ns = 20 QNSB
involves a full 2-oscillator product basis of Ns2 = 400 states, thus a 400 × 400 Hamiltonian
matrix with 3364 non-vanishing elements out of 160000 (see Fig. 6.1).
Distorted Morse potential for the individual oscillator, where in general QNSB performs
better than Morse basis, and the optimized QNSB+ can be straightforwardly used. Consider
now the same Hamiltonian, but with a 12 times stronger intermode potential coupling (C1,3 =
600, while the kinetic coupling strength is the same as in the previous calculation). Table 6.2
lists the resulting eigenergies, obtained via diagonalization on Ns = 10, Ns = 20 and Ns = 50
QNSB (involving 100 × 100, 400 × 400 and 2500 × 2500 matrices respectively). Here the
convergent trend of the eigenvalues, as Ns increases, is more evident. Moreover, in the Morse
basis (Ns = 10), only 20 bound states of the two-oscillator system are obtained. Increasing
the basis size to Ns = 20, the number of states below the dissociation threshold grows to 23.
The Ns = 50 calculation, which is probably well converged, finds 24 bound states. Clearly, for
strong intermode interaction, a larger basis is needed for a well converged description, even in a
particularly favorable situation of pure Morse individual stretching oscillators. The role of the
distortion from the Morse shape of the individual oscillators is also an important aspect of the
model to be investigated, and here the QNSB+ is likely to be very useful. More investigation
of the HCAO model and of extensions thereof within the present scheme is a necessary starting
step to extend the method to actual polyatomic molecules.
0
state ĤHCAO Ĥ (Ns = 10) Ĥ (Ns = 20)
0 -1125.702 -1125.578 -1125.578
1 -1011.165 -1014.568 -1014.569
2 -1011.165 -1007.072 -1007.072
3 -909.645 -911.564 -911.564
4 -909.645 -909.504 -909.504
5 -896.629 -892.709 -892.709
6 -821.139 -821.419 -821.419
7 -821.139 -821.277 -821.277
8 -795.108 -798.677 -798.679
9 -795.108 -786.243 -786.245
10 -745.650 -745.302 -745.303
11 -745.650 -745.253 -745.254
12 -706.603 -708.029 -708.041
13 -706.603 -704.887 -704.897
14 -693.587 -686.073 -686.077
15 -683.176 -682.184 -682.185
16 -683.176 -681.560 -681.562
17 -633.718 -634.390 -634.416
18 -633.718 -634.356 -634.381
19 -631.113 -627.669 -627.699
20 -631.113 -627.401 -627.431
21 -605.082 -607.140 -607.168
22 -605.082 -596.280 -596.289
23 -597.275 -596.225 -596.235
24 -597.275 -592.004 -592.024
25 -573.848 -572.929 -572.968
26 -573.848 -572.920 -572.958
27 -568.639 -564.761 -565.082
28 -568.639 -564.465 -564.791
29 -563.436 -561.747 -561.758
30 -563.436 -561.721 -561.736
Table 6.1. The second and third columns list the eigenvalues computed by
0 0
diagonalizing ĤHCAO and Ĥ = ĤHCAO + Ĥ1,3 respectively in the Morse basis (the
parameters are detailed in the text). The coupling has the effect of breaking
the degeneracy of Morse levels, introducing small splittings. Notice that while
0
Ĥ = ĤHCAO matrix is diagonal, all 10000 matrix elements in the complete Ĥ
matrix in the Morse basis are non-zero. The eigenvalues of the third column are
reproduced exactly using a QNSB (43) with the same number Ns = 10 of basis
states. In the QNSB the number of non-vanishing matrix elements is 784 only.
The fourth column refers to a QNSB of Ns = 20 states: for the small interbond
couplings considered here, the improved convergence shifts the eigenvalues only
slightly. The estimated dissociation energy here is Ediss = −559.673 according to
Eq. (191). The number of correct bound states is produced already using Morse
basis.
6.3. COMPARISONS 79
state Ĥ (Ns = 10) Ĥ (Ns = 20) Ĥ (Ns = 50)

0 -1127.953 -1127.962 -1127.962
1 -1044.329 -1044.469 -1044.469
2 -994.655 -994.807 -994.807
3 -950.574 -951.391 -951.391
4 -932.176 -932.749 -932.749
5 -882.690 -883.669 -883.669
6 -862.485 -865.327 -865.327
7 -837.397 -838.776 -838.776
8 -831.236 -834.098 -834.098
9 -798.924 -801.387 -801.387
10 -780.472 -783.576 -783.576
11 -755.336 -766.217 -766.217
12 -748.089 -751.867 -751.867
13 -745.651 -750.922 -750.922
14 -717.123 -723.132 -723.132
15 -715.594 -721.154 -721.154
16 -695.972 -701.923 -701.924
17 -691.685 -699.083 -699.086
18 -677.196 -687.456 -687.456
19 -676.347 -684.679 -684.691
20 -667.305 -678.441 -678.563
21 -666.900 -676.660 -676.663
22 -661.288 -674.724 -674.944
23 -654.906 -673.534 -674.758
Table 6.2. The eigenvalues computed diagonalizing a strongly coupled Ĥ =
0
ĤHCAO + Ĥ1,3 in the QNSB for three different basis sizes. The estimated disso-
ciation threshold for this potential energy surface is Ediss = −674.611, according
to Eq. (191). Energies above this threshold are marked by small characters. The
diagonalization within the pure Morse basis (Ns = 10) produces only 20 states
below this threshold, thus indicating rather poor convergence.
CHAPTER 7
Discussion and outlook
The present work addresses the problem of the systematic inclusion of anharmonic effects in
the computation of molecular vibrational spectra. This is a long-standing problem, and several
treatments have been proposed at least since the original Morse work [24]. It is evident that
an accurate Hamiltonian treatment of molecular systems requires an accurate knowledge and
a careful representation of the adiabatic molecular potential, and much effort was devoted to
devise clever potential representations. This leads to the identification of coordinate systems
which allow a synthetic, realistic and flexible functional form for the potential [5, 15, 24]. As
analytic matrix elements are usually unavailable for a general potential function, the choice of
the coordinate representation must be gauged also by the simplicity of the determination of
numerical solutions. From this point of view, harmonic-oscillator-based methods are probably
the most preferable. Nonetheless, the Morse potential shares at least two distinctive features
with the harmonic oscillator potential. The first is that both are 1D solvable potentials, so
both can be conveniently taken as good starting points for a more accurate approximation
of realistic potentials. The second common feature was recognized by means of dynamical
symmetries: the 1D Morse oscillator and (truncated) harmonic oscillator are represented by
two isomorphic algebras (O(2) and U (1) respectively) [40, 41].
Traditionally, harmonic-oscillator-based treatments have been pursued and developed the
most: satisfactory accuracy was achieved in 1D systems, at the expense of considering high-
order expansions in the displacement coordinate, mainly following perturbative approaches
[57, 58, 59].
Morse-coordinate-based methods were also proposed and employed for diatomic molecules,
such as the PMO method developed in the 1970’s [26, 27, 29, 30], as well as other methods [31],
and later the potential expansion in Morse coordinate was also introduced in the spectroscopy
of polyatomic molecules [2, 6, 9, 10, 11, 32, 49]. All these works face the problem of
the finiteness of the Morse basis, and its consequent intrinsic inadequacy to describe general
potentials. Moreover, the simplifications granted by the ladder-operators-based formulation
of harmonic oscillator problem is not available, and this makes these Morse-based models
substantially more intricate (non-sparse matrices).
The present work attempts to develop a method delivering at the same time a microscop-
ically accurate parameterization of the adiabatic PES and a manageable algebraic approach,
requiring minimal numerical efforts. The choice of Morse potential, or more precisely the
choice of Morse coordinate as a starting point is a natural one, following the relatively re-
cent algebraic treatments since the introduction of dynamical symmetries in molecular physics
[40, 41, 60, 61, 62]. Our work initially examined existing models and methods, with the aim
of finding some possible generalization within dynamical symmetries formalism, but after some
initial attempts, we found no fully satisfactory theory and decided to develop a new one based
on combining recent findings in the broad context of the Morse oscillator.
Eventually, we partly succeeded in our task. On one side, in Chaps. 3 and 4, we identify
a potential expansion in Morse coordinate (34) that is at the same time flexible and accurate:
with comparably few terms this expansion can approximate a wide class of typical potential
81
82 7. DISCUSSION AND OUTLOOK
accurately without introducing spurious unphysical behavior, a problem often overlooked in the
past, both for harmonic-oscillator-based and even for Morse-based expansions [3]. The PMO
prescription [26, 27, 29] deals with this problem, but the attention is mainly devoted to obtain
an accurate approximation at the minimum region.
In the potential expansion (34), the Morse term may find a comparably small role: this can
make the Morse-term basis not particularly accurate to describe the whole molecule spectrum.
Morse basis incompleteness becomes a critical problem. Our solution to this drawback is
the introduction of the QNSB [14, 35]. Following the instructive algebraic approach from
SUSY quantum mechanics, we introduce a Hilbert-space basis that for the first Ns terms
is equivalent to the Morse basis, but is complete. In this basis the Morse Hamiltonian is
tridiagonal, and the matrix elements of any power term in the potential expansion (34) can
be computed algebraically, and the resulting matrices are band-diagonal, while in Morse basis
they would be non-sparse [31].
By combining these two ingredients, the clever potential expansion and the complete QNSB,
virtually any well-behaved molecular 1D potential problem can be solved, without intrinsic
limitation in the achievable accuracy. In this way the Morse-coordinate expansion produces
a problem which can cast in a form strictly resembling harmonic-oscillator-based methods.
Chapter 4 reports a few examples of applications of this method [42], demonstrating a fairly
uniform accuracy in the computed spectrum throughout the whole energy region, from the
bottom of the potential well to dissociation. This accuracy is not restricted to potentials
especially Morse-like (see Section 2.5.7). This method has therefore a good predictive power,
whose accuracy is basically limited by the accuracy of the available PES data.
Other methods involve clever choices of expansion variables for diatomics, but with an en-
tirely different starting point. For example, the direct-potential-fit methods [19, 20, 21, 22,
46, 47, 48] take the experimental spectra as input: the validity of the potential parameter-
ization generated by such methods is usually restricted to the energy range where data are
available, and the predictive power is doubtful. The method of the present work generates
instead purely vibrational spectra based on first-principle determination of the adiabatic PES,
thus with relatively constant accuracy through the whole bound-state energy range. It would
be difficult (and of little interest) to attempt to use the present formalism to fit experimen-
tal lines, as in the direct-potential-fit approach. An extension to address non-adiabatic and
rovibrational effects could be developed in future work.
When this method is extended to a polyatomic context, the total basis size and the Hamil-
tonian sparseness are the main computational issues to consider. If in the present scheme
spectroscopic accuracy can be obtained with as few as 30 states per degree of freedom, the
total basis size could still approach 306 ' 109 states for a tetraatomic molecule. This size
could exceed the computational power available, even if the matrix to diagonalize is sparse.
As common practice in quantum chemical calculations, a truncated basis could be selected by
accepting slightly less accurate spectra. This truncation may target specific 1D oscillators: in
this case, when Ns is reduced to the number [s] + 1 of the underlying Morse term, the QNSB
and the Morse basis are equivalent and yield equivalent spectra. If further reduction is needed,
the convergence of the two resulting spectra shows a characteristically different deterioration.
Truncation of the Morse basis affects greatly the high-energy states near dissociation: these
disappear rapidly one by one, without significant modification of the low-energy levels. On
the contrary, the reduction of QNSB size affects eigenenergies throughout the whole spectrum,
but with a much slower decrease of the number of bound states. Accordingly, the two basis
can be both used with success, keeping in mind their different behavior in different situations.
7. DISCUSSION AND OUTLOOK 83
Alternatively, standard truncation of the full many-oscillator basis, based on the relative per-
turbative weight of the states [3, 63] should also be efficient due to the very sparse structure
of the Hamiltonian matrix elements.
To take care of efficiency and basis size constraints related to computational resources, we
also discussed in some detail the possibility of a more clever QNSB, the QNSB+, optimized
to make convergency of the computation fastest. Though we could not find a simple algebraic
prescription for the choice of the QNSB+ for given Hamiltonian, the analysis of Chap. 5 pro-
poses a fairly simple empirical procedure to determine the optimal QNSB+, thus obtaining a
substantial basis-size reduction for the same variational accuracy.
This last discussion could be very useful in the extension of the present method to compute
vibrational spectra of polyatomic molecules. Due to the similarities outlined above, it is not
especially more complicated than the formally similar traditional harmonic-oscillator based
expansion. In particular it lends itself naturally to both diagonalization (Lanczos / Davidson)
[64, 65, 66, 67] and Green-function–based [3, 68] iterative methods of solution. Special care
must be put in making sure that the total polyatomic adiabatic potential surface (e.g. fitted
on ab initio data) has a single absolute minimum, but eventually this global property is easier
to implement than on traditional Taylor power-series methods where the global shape of the
approximate potential is dictated by local properties at the equilibrium geometry. A separate
treatment of “soft” torsional coordinates may be needed, exactly like within the traditional
normal-coordinate harmonic-oscillator based schemes.
Chapter 6, only starts to address the polyatomic problem: preliminary work ckecks if the
QNSB replaces efficiently the standard Morse basis expansion in multi-mode contexts. In
polyatomic systems the overall accuracy of any technique is the combined result of the accuracy
of the treatment of several different aspects, including in particular the kinetic term: our
methods presently can only deal well with the potential features. All the same, several currently
used models employ largely Morse related quantities even in treating polyatomic molecules
[6, 9, 10, 11, 49]. In all these models, our method can give at least a contribution.
Eventually, to address the full molecular problem it will be useful to develop a similar
method for other relevant potentials. For example, the Pöschl-Teller potential [69] could be
at the basis of the description of bending vibrations in polyatomic molecules. Like the Morse
potential, this is also a solvable potential, so an approach similar to that followed in this work
could be attempted, and become another useful ingredient to improve the accuracy and man-
ageability of present-day theoretical models. At last, however, the main difficulty to overcome
is the task of dealing with the intricacies of the kinetic term in general coordinates within a
QNSB.
CHAPTER 8
Programs and codes
To implement and use in practice the formalism of previous chapters, we have developed
a number of simple Mathematica codes. The Mathematica programming environment is a
powerful tool for formal manipulations, wich makes a wide library of mathematical functions
available. In particular, Morse related eigenfunctions and eigenvalues computation uses exten-
sively the definition and relations among the generalized Laguerre polynomials, as well as of
customary calculus. These reasons made us choose Mathematica as a development environ-
ment, though it is not intended as the environment where the final routine computations should
be carried out. As a collateral observation, we want to stress that even if the huge quantity of
sophisticated tools that Mathematica provides is surely very useful, we recommend not to trust
blindly their correctness, as we came across several bugs of the interpreter and varied behavior
of different versions.
We describe here briefly the Mathematica codes developed in the course of our work:
• MorseMatEl.m
Produce the matrix elements in Morse basis for the solution by exact diagonal-
ization of a general 1D Schrödinger equation for a potential of the form of expan-
sion (34). Input Variables: Morse parameters (x0 , ~, M, P aram), where P aram =
{V0 , a3 , .., aNmax , α} contains the expansion coefficients and α. The diagonal Morse
Hamiltonian is computed using exact eigenvalues’ expression (29). The [v(x)]i matrix
elements are computed via Laguerre Polynomial integral relations (21). These func-
tions work also for non integer i ≥ 0. If not specified, Ns = [s]+1, s as in Eq. (24). This
code also computes automatically QNSB (s, σ), necessary for internal use. Optional
by specifying Ns , one can set the number of basis states to any Ns ≤ [s] + 1.
• MorseOscill1d.m
Produce the solution by exact diagonalization in the Morse basis of a general 1D
Schrödinger equation for a potential of the form of expansion (34). Input Variables:
Morse parameters (x0 , ~, M, P aram), where P aram = {V0 , a3 , .., aNmax , α} contains the
expansion coefficients and α.
Needs: MorseMatEl.m .
• MatElExact.m
Routine for the exact computation of the general form of the matrix elements
of [e−α(x−x0 ) ]i , [v(x)]i , in the QNSB (43), e.g. those listed in Appendix 9.4. The
algorithm is based on the algebraic form (91) of the above operators. Due to its
algebraic nature, the exponent i is restricted to integer values (another code based
on analytic Laguerre Polynomial integral relations (21) is available, but it has no
evident utility). This method is very time consuming, due to the necessity to manage
all the formulas symbolically, thus we also developed an approximate routine to be
employed in standard applications, in particular on slow computers. This is contained
in MatElGen.m .
• MatElGen.m
85
86 8. PROGRAMS AND CODES
Fast general routine for the computation of the matrix elements of a Morse power
expansion (34), according to the generalized formulation of Chap. 51.
Input Variables: Morse expansion (34) parameters: (x0 , ~, M, P aram) where P aram =
{V0 , a3 , .., aNmax , α} as before. By default, Ns = [s] + 1 (s is defined by Eq. (24) and
σ is defined by Eq. (96)), produces the QNSB of Eq. (43) and all matrix elements
are computed on the regular QNSB. Optional: by (sB = esse, σB = sigorig, Ns )
(Ns = number of basis states) can be specified: in this case all matrix elements are
computed in the corresponding QNSB+. If a reference set of eigenvalues eigRef is
provided as additional input vector, it can automatically produce the square sum dif-
ference eigDiff, which can be employed in a minimization, for example to determine
the QNSB+.
• BasisfuncGen.m
Definitions of normalized Morse bound states and QNSB (in their generalized form).
Input values are:
– Morse parameters (V0 , α, x0 ); QNSB parameters by default are automatically ob-
tained from the Morse ones.
– Optionally, the generalized QNSB parameters (sB , σB , x0B , αB ).
– Optionally, it can read a table with numerical wavefunctions produced by external
programs to compute the projections of those numerically computed eigenfunc-
tions on Morse basis and on (generalized) QNSB.
Also generates projections of numerically computed eigenfunctions on Morse basis
and on QNSB/QNSB+.
• FitGener.m
Compute the optimal parameters in the 1D potential expansion (34). Input pa-
rameters: the degree of expansion degExp, and optionally the position of the minimum
xmin and the well depth welldpt. The target potential is usually read from an input
file Ipot.dat. Some tuning is possible, for example by introducing external constraints.
• HCAOGen.m
Implementation of the standard HCAO model and of its QNSB counterpart. In-
put parameters: the Morse potential parameters for two oscillators, P aram1,2 , their
expansion coefficients C1,3 = c13 and the intermode coupling cos α = cosalfa (see
Chapt. 6).
Needs: MatElGen.m (or MatElExact.m), MorseOscill1d.m.
All the listed pieces of codes are integrated with one another seamlessly. As a practical
example, to execute the calculation producing the spectrum in the right column of Table 2.14,
one executes the following code at the Mathematica prompt:
x0=1;
hbar=1;
M=1;
Param={625,0,0,0,625,4};
NS=40;
<< MatElGen.m
eigvals=eigenvalues[matHam[s,sig,NS]]//TableForm
1The main difference with respect to the exact method is that the expression for the general matrix element
is not computed a priori. This routine produces the non-vanishing elements of the given row of the target
matrix, one by one. This leads to some numerical noise, that for very high-order expansion potential (34), must
be taken into account.
CHAPTER 9
Appendixes
9.1. Appendix A: Laguerre polynomial integrals

The explicit form of Morse matrix elements involves the generalized Laguerre polynomials
integral relation:
Z ∞
µ0
X λ − µλ − µ0 λ + k
λ −z µ n+n0
(192) z e Ln (z)Ln0 (z)dz = (−) Γ(λ + 1) .
0 n−k n0 − k k
k
This equality remains valid for all λ satisfying Re(λ) > −1. It allows to compute the scalar
products involving Morse eigenvectors, and the matrix elements of operators of the type ẑ t ,
ẑ ∝ (e−α(x̂−x0 ) ) in the Morse basis.
Tables 9.1, 9.2 collect the vanishing condition for the
n
generalized binomial coefficient m , appearing in the result of integrals of the kind of Eq. (192).
As a first example of application of Eq. (192) we verify that Morse eigenfunctions are
orthonormal. We have
0 µ−µ0 µ+µ0
n+m µ + µ0 X µ 2−µ − 1 2
−1 2
−1+s
(193) hn|mi = Nn Nm (−) Γ( ) .
2 s
n − s m − s s
We specialize the Eq. (193) to Morse eigenstates, i.e. µ = k − 2n − 1, µ0 = k − 2m − 1,

0
λ = µ+µ
2
− 1. In particular, µ − µ0 = 2(m − n), µ0 − µ = 2(n − m), µ + µ0 = 2k − 4(n + m) − 2.
Accordingly, the scalar product of two Morse wavefunctions is
(194)
X n − m − 1m − n − 1k − (n + m) − 2 + s
n+m
hn|mi = Nn Nm (−) Γ(k−(n+m)−1) .
s
n − s m − s s
n

m
<0 =0 >0
<0 0 0 0
=0 1 1 1
> 0 6= 0 0 6= 0 if m ≤ n, and 0 if m > n
n
Table 9.1. Generalized binomial coefficients m vanishing condition, n, m rel-
ative integers.
r

<0 =0 >0
m
<0 0 0 0
=0 1 1 1
> 0 6= 0 6= 0 6=0
Table 9.2. Generalized binomial coefficients mr vanishing condition, r real non
integer, m relative integer.
87
88 9. APPENDIXES
If n = m, we have
X −1 2 k − 2n − 2 + s
(195) hn|ni = Nn2 Γ(k − 2n − 1) .
s
n − s s
From Table 9.1, the only non-vanishing terms in the sum are those satisfying k ≤ n, i.e.
n 2
2
X −1 k − 2n − 2 + s
(196) hn|ni = Nn Γ(k − 2n − 1) ,
s
n − s s
n
X Γ(k − 2n − 2 + s + 1)
= Nn2 Γ(k − 2n − 1) = 1.
s
Γ(k − 2n − 2 + 1)Γ(s + 1)
If instead n 6= m, imposing n = m + d, d ≥ 1, we can write
X d − 1−d − 1k − 2m − d − 2 + s
d
(197) hn|mi = (−) Nn Nm Γ(k − 2m − d − 1) .
s
n−s m−s s
If d > 1 the first binomial is non-vanishing only if m + 1 ≤ s ≤ n, and the second
binomial is
0
non-zero for s ≤ m, thus the sum vanishes. If d = 1, the first binomial is n−s , and is non-zero
−2

only for s ≡ n, and the second binomial reads m−s , non-zero only for s ≤ m, thus the sum
again vanishes, as it should. We conclude that
(198) hn|mi = δnm .
Next consider matrix elements of ẑ = ke−α(x̂−x0 )
(199) hn|ẑ t |mi .
These expressions involve integrals of the general form (192), namely:
Z ∞
0
z λ+t e−z Lµn (z)Lµn0 (z)dz = (−)n+n Γ(λ + t + 1)
0
0
X λ + t − µλ + t − µ0 λ + t + k
(200) .
n−k n0 − k k
k
Assuming again n = m + d, d ≥ 1, we have:

hn|ẑ t |mi = (−)d Nn Nm Γ(k − 2m − d − 1 + t) ×
X d − 1 + t−d − 1 + tk − 2m − d − 2 + t + s
(201) .
s
n−s m−s s
Let us distinguish two cases: if t is an integer, then
<0if d − 1 + t > 0 (i.e. d > 1 − t), then
>0
−d − 1 + t < 0, and this is a case of the form n−s m−s , i.e. we have:
m
X d − 1 + t −d − 1 + t k − 2m − d − 2 + t + s
(202) .
s=m+1−t
n−s m−s s
If instead t is not an integer, we have:
min(n,m)
X d − 1 + t −d − 1 + t k − 2m − d − 2 + t + s
(203) .
s
n − s m − s s
9.2. APPENDIX B: MORSE 1D AND 2D 89
9.2. Appendix B: Morse 1D and 2D

The Morse potential allows eigenfunctions of the same form of the radial part of the 3D
Schrödinger equation also in 2D and 1D. For sake of completeness, we report here also the
derivation for these two cases.
Consider the 1D Schrödinger equation for the Morse potential:
2 2
~ d −α(x−x0 ) −2α(x−x0 )

(204) − + V0 −2e +e Ψ(x) = W Ψ(x) .
2µ dx2
The x variable is defined in (−∞, +∞).
With the change in coordinate y = e−α(x−x0 ) , the equation become:
d d2 2 2 d
2
d
(205) = −ay, 2
= a y 2
+ a2 y ,
dx dx dy dy
and thus to

d2 Ψ 1 dΨ 2µ W 2V0
(206) + + 2 2 + − V0 Ψ = 0 ,
dy 2 y dy ~ a y2 y
that is the same as Eq. (9). The eigenfunction is exactly the same i.e.:
k −α(x−x0 ) k−1−2n
(207) Ψn (x) = e− 2 e (ke−α(x−x0 ) ) 2 Lk−1−2n
n (ke−α(x−x0 ) ) ,
and k is defined by Eq. (16). The normalization condition is instead different from those of 3D
solution. The normalization condition hΨn |Ψm i = δnm reads
Z +∞
−α(x−x0 )
(208) e−ke (ke−α(x−x0 ) )k−n−m−1 Lk−1−2n
n (ke−α(x−x0 ) )Lk−1−2m
m (ke−α(x−x0 ) ) = δn,m .
−∞
Upon the coordinate transformation y = ke−α(x−x0 ) , neglecting a proportionality factor, we

have:
Z +∞
(209) e−y (y)k−n−m−2 Lk−1−2n
n (y)Lk−1−2m
m (y) ,
0
and this is not exactly the radial part of Eq. (20).
In 2D, in polar coordinates the Schrödinger equation for the Morse potential is:
2 2
~ ∇ −α(ρ−ρ0 ) −2α(ρ−ρ0 )

(210) + V0 −2e +e Ψ(ρ) = W Ψ(ρ) ,
2µ dρ2
2
1 ∂ ∂
2
where ∇ ≡ ρ ∂ρ ρ ∂ρ + ρ12 φ∂ 2 .
Applying the usual 2D separation of variable ansatz,
χm (ρ) eimφ
(211) Ψ(ρ, φ) ≡ √ ,
ρ1/2 2π
we obtain that χm must satisfy the following equation:
2 2 2
~ d 2 1 ~ −α(ρ−ρ0 ) −2α(ρ−ρ0 )

(212) − + m + + V0 −2e +e − W χm (ρ) = 0 .
2µ dρ2 4 2µρ2
~2
Neglecting the rotational term m2 + 41 2µρ 2 , the usual relation 9.2 gives:

d2 χ 1 dχ 2µ W 2V0
(213) + + + − V0 χ = 0 .
dy 2 y dy ~2 a2 y 2 y
This equation is the same as the radial part of Eq. (20).
90 9. APPENDIXES
The complete eigenfunction thus becomes

k −α(ρ−ρ0 ) k−1−2n eimφ
(214) Ψ(ρ, φ) = e− 2 e (ke−α(ρ−ρ0 ) ) 2 Lk−1−2n
n (ke−α(ρ−ρ0 ) ) √ .
2πρ1/2
The normalization condition is then:
(215)
Z Z
dφ ∞ −α(ρ−ρ0 ) 1
ρdρe−ke (ke−α(ρ−ρ0 ) )k−n−m−1 Lk−1−2n
n (ke−α(ρ−ρ0 ) ) Lm
k−1−2m
(ke−α(ρ−ρ0 ) ) =
2π 0 ρ
Z ∞
−α(ρ−ρ0 )
= e−ke (ke−α(ρ−ρ0 ) )k−n−m−1 Lk−1−2n
n (ke−α(ρ−ρ0 ) )Lk−1−2m
m (ke−α(ρ−ρ0 ) ) = δn,m .
0
The coordinate transformation y = ke−α(ρ−ρ0 ) , gives
Z keaρ0
(216) e−y (y)k−n−m−2 Lk−1−2n
n (y)Lk−1−2m
m (y)dy .
0
This coincides with the radial normalization of the 3D case, Eq. (20).
9.3. Appendix C: Morse expansion derivatives

Every term of the form (e−α(x−x0 ) − 1)n has in the minimum vanishing k th derivative for
0 ≤ k < n, and non-vanishing for k ≥ n. The general form of the k th derivative, with k ≥ n, is
dk
k
v(x)n = (−1)n [c1 e−α(x−x0 ) (1 − e−α(x−x0 ) )n−1
dx
+c2 e−2α(x−x0 ) (1 − e−α(x−x0 ) )n−2 + ... + cn e−nα(x−x0 ) ]
Xn
(217) ≡ (−1)n ci e−iα(x−x0 ) (1 − e−α(x−x0 ) )n−i .
i=1
The derivatives of any order share the same form, with coefficients c0i related by the recursive
relation:
(218) ck+1
1 = ck1 ,
(219) ck+1
i = (n − i + 1)cki−1 − icki , i = 2, ..., n .
The derivative of order k is a sum of a number of k terms (if k < n), namely:
k i
dk i
XY
(220) k
v(x) = (n − l + 1)di e−iα(x−x0 ) (1 − e−α(x−x0 ) )n−i .
dx i=1 l=1
The numerical coefficients dki , listed in Table 9.3, are obtained recursively from
(221) dki = k · dki−1 + di−1
k−1
, d11 = 1 .
9.4. Appendix D: Matrix elements in the QNSB

We report here the nonzero matrix elements of the cubic term [v̂(x)]3 in the QNSB:
Cn−2 Cn−1 Cn
(222) hφn |[v̂(x)]3 |φn−3 i = −
(1 + 2s)3
3Cn−1 Cn (2n − 2[s] − 3)

(223) hφn |[v̂(x)]3 |φn−2 i =
(1 + 2s)3
9.4. APPENDIX D: MATRIX ELEMENTS IN THE QNSB 91
i
1 2
k 3 4 5 6 7 8 9
1 1 0 0 0 0 0 0 0 0
2 1 1 0 0 0 0 0 0 0
3 1 3 1 0 0 0 0 0 0
4 1 7 6 1 0 0 0 0 0
5 1 15 25 10 1 0 0 0 0
6 1 31 90 65 15 1 0 0 0
7 1 63 301 350 140 21 1 0 0
8 1 127 966 1701 1050 266 28 1 0
9 1 255 3025 7770 6951 2646 462 36 1
k
Table 9.3. The di coefficients appearing in Eq. (220) for the derivative of order
k of v(x)i .
Cn
(224) hφn |[v̂(x)]3 |φn−1 i = − (1+2s) 2 2 2 2
3 (12n − 24n + 13 + Cn−1 + Cn + Cn+1
+12[s]2 − 24n[s])
1
(225) hφn |[v̂(x)]3 |φn i = (1+2s)3
[Cn2 (6n 2
− 6[s] − 5) + Cn+1 (6n − 6[s] − 1)
+(2n − 2[s] − 1)3 ].
Of course, hφn |[v̂(x)]3 |φn+3 i, hφn |[v̂(x)]3 |φn+2 i, ..., are obtained by Hermiticity.
Similarly, the nonzero matrix elements of the quartic term [v̂(x)]4 read:
Cn−3 Cn−2 Cn−1 Cn
(226) hφn |[v̂(x)]4 |φn−4 i =
(1 + 2s)4
8Cn−2 Cn−1 Cn (2 − n + [s])

(227) hφn |[v̂(x)]4 |φn−3 i =
(1 + 2s)4
Cn−1 Cn
(228) hφn |[v̂(x)]4 |φn−2 i = (1+2s)4
(58 − 72n + 24n2 + Cn−2
2 2
+ Cn−1 + Cn2
2
+Cn+1 + 72[s] − 48n[s] + 24[s]2 )
4Cn
(229) hφn |[v̂(x)]4 |φn−1 i = − (1+2s) 2 3 2
4 [−10 + 26n − 24n + 8n − Cn+1
2 2
+2nCn+1 + Cn−1 (2n − 3 − 2[s]) + 2Cn2 (n − 1 − [s]) − 26[s] + 48n[s]
−24n2 [s] − 2Cn+1
2
[s] − 24[s]2 + 24n[s]2 − 8[s]3 ]
1
(230) hφn |[v̂(x)]4 |φn i =
(1+2s)4
{Cn4 + Cn+1
4
+ (1 − 2n + 2[s])4 + Cn+12
×
(2 − 8n + 24n2 + Cn+2
2
+ (8 − 48n)[s] + 24[s]2 ) + Cn2 [Cn−1
2
+2(9 − 20n + 12n2 + Cn+1 2

+ 20[s] − 24n[s] + 12[s]2 )]}.
The nonzero matrix elements of the term [v̂(x)]5 read:
Cn−4 Cn−3 Cn−2 Cn−1 Cn
(231) hφn |[v̂(x)]5 |φn−5 i = −
(1 + 2s)5
5Cn−3 Cn−2 Cn−1 Cn

(232) hφn |[v̂(x)]5 |φn−4 i = − (2n − 5 − 2[s])
(1 + 2s)5
92 9. APPENDIXES
(233) hφn |[v̂(x)]5 |φn−3 i = − Cn−2 Cn−1 Cn

(1+2s)5
(170 − 160n + 40n2 + Cn−3
2
2 2
+Cn−2 + Cn−1 + Cn2 + Cn+1
2
+ 160[s] − 80n[s] + 40[s]2 )
(234) hφn |[v̂(x)]5 |φn−2 i = − C(1+2s)

n−1 Cn 2 3 2
5 [330 − 580n + 360n − 80n + 13Cn
−10nCn2 + 9Cn+1
2 2
− 10nCn+1 + 580[s] − 720n[s] + 240n2 [s] + 10Cn2 [s]
2
+10Cn+1 [s] + 360[s]2 − 240n[s]2 + 80[s]3 + Cn−1
2
(17 − 10n + 10[s])
2
+Cn−2 (21 − 10n + 10[s])]
Cn
(235) hφn |[v̂(x)]5 |φn−1 i = − (1+2s) 2 3 4 4
5 [121 − 400n + 520n − 320n + 80n + Cn−1
+Cn4 + 14Cn+1
2 2
− 40nCn+1 + 40n2 Cn+12 4
+ Cn+1 2
+ Cn+1 2
Cn+2 + 400[s]
−1040n[s] + 960n [s] − 320n [s] + 40Cn+1 [s] − 80nCn+1 [s] + 520[s]2
2 3 2 2
−960n[s]2 + 480n2 [s]2 + 40Cn+12

[s]2 + 320[s]3 − 320n[s]3 + 80[s]4
+2Cn2 (21 − 40n + 20n2 + Cn+1
2
− 40(n − 1)[s] + 20[s]2 ) + Cn−1
2
(94
2 2 2 2 2
−120n + 40n + Cn−2 + 2Cn + Cn+1 + 120[s] − 80n[s] + 40[s] )]
1
(236) hφn |[v̂(x)]5 |φn i = − (1+2s) 4
5 {Cn+1 (1 − 10n + 10[s])
+Cn4 (9 − 10n + 10[s]) + Cn+1

2
[2 − 20n + 40n2 − 80n3 + 20(1 − 4n
+12n2 )[s] − 40(6n − 1)[s]2 + 80[s]3 + Cn+2
2
(10[s] − 3 − 10n)]
+Cn {Cn−1 (13 − 10n + 10[s]) + 2[29 − 90n + 100n2 − 40n3
2 2
+10(9 − 20n + 12n2 )[s] − 20(6n − 5)[s]2 + 40[s]3

2
+Cn+1 (5 − 10n + 10[s])]} + (1 − 2n + 2[s])5 }.
The nonzero matrix elements of the term [v̂(x)]6 read:
Cn−5 Cn−4 Cn−3 Cn−2 Cn−1 Cn

(237) hφn |[v̂(x)]6 |φn−6 i =
(1 + 2s)6
12Cn−4 Cn−3 Cn−2 Cn−1 Cn

(238) hφn |[v̂(x)]6 |φn−5 i = (3 − n + [s])
(1 + 2s)6
Cn−3 Cn−2 Cn−1 Cn

(239) hφn |[v̂(x)]6 |φn−4 i = (1+2s)6
(395 − 300n + 60n2 + Cn−4
2
2 2 2
+Cn−3 + Cn−2 + Cn−1 + Cn2 + Cn+1
2
+ 300[s] − 120n[s] + 60[s]2 )
(240) hφn |[v̂(x)]6 |φn−3 i = − 4Cn−2 Cn−1 Cn

(1+2s)6
[−380 + 510n − 240n2 + 40n3
2 2
−6Cn−1 + 3nCn−1 − 5Cn2 + 3nCn2 − 4Cn+12
+ 3nCn+12 2
+ Cn−3 ×
2
(−8 + 3n − 3[s]) + Cn−2 (−7 + 3n − 3[s]) − 510[s] + 480n[s]
2 2
−120n [s] − 3Cn−1 [s] − 3Cn2 [s] − 3Cn+1
2
[s] − 240[s]2
+120n[s]2 − 40[s]3 ]
9.6. APPENDIX F : DYNAMICAL SYMMETRY APPROACH 93
Cn−1 Cn
(241) hφn |[v̂(x)]6 |φn−2 i =
(1+2s)6
[1771 − 3960n + 3480n2 − 1440n3 + 240n4
4 4
+Cn−2 + Cn−1 + 107Cn2 − 156nCn2 + 60n2 Cn2 + Cn4 + 59Cn+1 2 2
− 108nCn+1
+60n2 Cn+12
+ 2Cn2 Cn+1
2 4
+ Cn+1 2
+ Cn+1 2
Cn+2 + 3960[s] − 6960n[s]
2 3 2 2 2 2
+4320n [s] − 960n [s] + 156Cn [s] − 120nCn [s] + 108Cn+1 [s] − 120nCn+1 [s]
+3480[s]2 − 4320n[s]2 + 1440n2 [s]2 + 60Cn2 [s]2 + 60Cn+12
[s]2 + 1440[s]3
−960n[s]3 + 240[s]4 + Cn−12
(179 − 204n + 60n2 + 2Cn2 + Cn+1 2
+ 204[s]
−120n[s] + 60[s]2 ) + Cn−2
2
(275 − 252n + 60n2 + Cn−3
2
+ 2Cn−1 2
+ Cn2
2
+Cn+1 + 252[s] − 120n[s] + 60[s]2 )]
4Cn
(242) hφn |[v̂(x)]6 |φn−1 i = − (1+2s) 2 3 4
6 [−91 + 363n − 600n + 520n − 240n + 48n
5
2 2
−11Cn+1 + 42nCn+1 − 60n2 Cn+1
2
+ 40n3 Cn+1
2 4
− Cn+1 4
+ 3nCn+1
2 2 4
+3nCn+1 Cn+2 + Cn−1 (−5 + 3n − 3[s]) + 3Cn4 (n − 1 − [s]) − 363[s]
+1200n[s] − 1560n2 [s] + 960n3 [s] − 240n4 [s] − 42Cn+1 2 2
[s] + 120nCn+1 [s]
−120n2 Cn+1
2 4
[s] − 3Cn+1 [s] − 3Cn+12 2
Cn+2 [s] − 600[s]2 + 1560n[s]2 − 1440n2 [s]2
+480n3 [s]2 − 60Cn+12
[s]2 + 120nCn+12
[s]2 − 520[s]3 + 960n[s]3
−480n2 [s]3 − 40Cn+12
[s]3 − 240[s]4 + 240n[s]4 − 48[s]5 + Cn−1 2
×
2 3 2 2 2
[−153 + 282n − 180n + 40n − 3Cn+1 + 3nCn+1 + 2Cn (−4 + 3n − 3[s])
2
+3Cn−2 (n − 2 − [s]) − 282[s] + 360n[s] − 120n2 [s] − 3Cn+1 2
[s] − 180[s]2 + 120n[s]2
−40[s]3 ] + 2Cn2 (−23 + 63n − 60n2 + 20n3 + Cn+1 2
(−2 + 3n − 3[s])
−3(21 − 40n + 20n )[s] + 60(n − 1)[s] − 20[s]3 )]
2 2
1
(243) hφn |[v̂(x)]6 |φn i = (1+2s)6
{Cn6 6
+ Cn+1 + (1 − 2n + 2[s])6 + Cn+1
4 2
{2Cn+2
+3[1 − 4n + 20n2 + (4 − 40n)[s] + 20[s]2 ]} + Cn4 [2Cn−1
2
+ 3(17 − 36n + 20n2 + Cn+1
2
+(36 − 40n)[s] + 20[s]2 )] + Cn2 {179 − 696n + 1080n2 − 800n3 + 240n4 + Cn−1 4
4
+3Cn+1 + 696[s] − 2160n[s] + 2400n2 [s] − 960n3 [s] + 1080[s]2 − 2400n[s]2
+1440n2 [s]2 + 800[s]3 − 960n[s]3 + 240[s]4 + Cn−12
(107 − 156n + 60n2
2 2
+Cn−2 + 2Cn+1 + 156[s] − 120n[s] + 60[s]2 ) + 2Cn+1
2
[19 − 60n + 60n2
2
+Cn+2 − 60(2n − 1)[s] + 60[s]2 ]} + Cn+1
2
{3 − 24n + 120n2 − 160n3 + 240n4
4
+Cn+2 + 24(1 − 10n + 20n2 − 40n3 )[s] + 120(1 − 4n + 12n2 )[s]2 − 160(6n − 1)[s]3
+240[s]4 + Cn+2
2
[11 + 36n + 60n2 + Cn+3 2
− 12(3 + 10n)[s] + 60[s]2 ]}}.
9.5. Appendix E: General properties of matrix elements

Assuming Morse potential parameters (V0 , α, x0 ), and considering the Morse basis (26) and
QNSB (43), we can draw general conclusions on many matrix elements of several relevant
operators. We collect several of these conditions in Table 9.4.
9.6. Appendix F : Dynamical Symmetry approach

Algebraic models usually allow a deeper understanding of the underlying principles of phys-
ical problems. One of the most appealing approach to molecular problems is probably that of
dynamical symmetries [40, 41, 61, 62].
94 9. APPENDIXES
Term Morse basis QNSB

−α(x−x0 ) k
hm|[e ] |ni ∀m, n iff |n − m| ≤ 2t + 1 (algebraic)
−α(x−x0 ) r
hm|[e ] |ni ∀m, n ∀m, n (analytic)
hm|[e−β(x−x0 ) ]|ni, β = lα ∀m, n iff |n − m| ≤ 2l + 1 (algebraic)
hm|[e−β(x−x0 ) ]|ni, β 6= lα ∀m, n ∀m, n (analytic)
hm|[e−α(x−x1 ) ]k |ni ∀m, n iff |n − m| ≤ 2t + 1 (algebraic)
−α(x−x1 ) r
hm|[e ] |ni ∀m, n ∀m, n (analytic)
−α(x−x0 ) k
hm|[e ] |ni ∀m, n iff |n − m| ≤ 2t + 1 (algebraic)
d
hm| dx |ni iff |n − m| = 1 iff |n − m| = 1 (algebraic)
dk
hm| dx k |ni iff |n − m| ≤ 2k + 1 iff |n − m| ≤ 2k + 1 (algebraic)
hm|ĤM |ni iff n = m iff |n − m| ≤ 5(3) (algebraic)
Table 9.4. Conditions for the matrix elements of relevant operators in the
Morse basis and QNSB to be non-zero. The parameters k and r are integer and
real respectively, and algebraic/analytic refer to the availability of expressions
to compute the matrix elements exploiting ladder operator relations (algebraic)
or through Laguerre polynomial integrals (Eq. (192)) (analytic). The matrix
elements in the Morse basis are all analytic, in this convention.
In a nutshell, the familiar realization of quantum mechanics is given in terms of differential

operators acting on the space of wavefunctions, as commonly done in Schrödinger wave me-
chanics. By means of creation and annihilation operators, an algebraic realization can be as
well given. Problems without spin, are formulated in terms of boson creation and annihilation
operators b†γ , bγ , satisfying the commutation relations:
(244) [bγ , b†γ 0 ] = δγγ 0 , [bγ , bγ 0 ] = [b†γ , b†γ 0 ] = 0 , γ, γ 0 = 1, ..., n + 1 ,
where n is the number of spatial degrees of freedom. All the operators of physical interest can
be written as power series in the bilinear products b†γ bδ . For example, the Hamiltonian Ĥ reads:
X 1 X
(245) Ĥ = E0 + γδ b†γ bδ + uγδγ 0 δ0 b†γ b†δ bδ bγ 0 bδ0 + ... ,
γδ
2 γδγ 0 δ0
and the number operator
X
(246) N̂ = b†γ bγ ,
γ
with no special physical meaning, but representing the total number of bosons, which commutes
with the Hamiltonian. The coefficients in expansion (245) can be related to the matrix elements
of the operators.
In this realization, the states of the system are written as Fock states:
1 † †
(247) |`i = b bγ 0 ... |0i ,
N` |γ {z }
` times
N` suitable normalization constants. Noting that bilinear products of creation and annihilation
operators
(248) Gγη = b†γ bη , γ, η = 1, ..., n + 1 ,
satisfy the commutation relations
(249) [Gγη , Gγ 0 η0 ] = Gγη0 δηγ − Gγ 0 η δη0 γ ,
we can say that they constitute the elements of the unitary algebra U (n+1): quantum mechanics
in n dimensions can be formulated in terms of the unitary algebra U (n + 1). This allows us to
write the Hamiltonian (245) in terms of the unitary algebra elements as
X 1 X 0
(250) Ĥ = E0 + 0γη Gγη + uγηγ 0 η0 Gγη Gγ 0 η0 + ... .
γη
2 0 0 γηγ η
The Hamiltonian (250) represents the general expansion in terms of the elements Gγη : it
corresponds to a Schrödinger equation with a generic potential. When the expansion (250)
involves only invariant Casimir operators (operators commuting with every operator of the
algebra) of the algebra G and of its subalgebras forming a “chain”, i.e.
(251) Ĥ = E0 + a1 C + a2 C 0 + a3 C 00 + ... ,
the Hamiltonian is said to possess a dynamical symmetry. This is a simple case, since all
Casimir operators are diagonal in a basis of G, so the spectrum can be obtained algebraically.
It turns out that dynamic symmetries correspond to solvable potentials in Schrödinger picture.
The 1D Morse potential and the (truncated) harmonic oscillator potential are both solvable
potentials, and both correspond to the unitary algebra U (2). Its standard realization is usually
given in terms of operators σ † , τ † , σ, τ . We will not address this problem in detail. We only
stress that they correspond to two different subalgebra chains of U (2):
(252) U (2) ⊂ U (1) harmonic oscillator,
(253) U (2) ⊂ O(2) Morse oscillator.
An interpretation of the algebraic Hamiltonian in the customary Schrödinger picture is neces-
sary, since general physical problems are usually defined in this last form. A geometric inter-
pretation of the algebraic Hamiltonian can be given constructing its first quantized version, and
this can be done associating a differential form to algebraic creation-annihilation operators:

i 0 ∂ † i 0 ∂
(254) τ = √ x + 0 , τ = −√ x − 0 ,
2 ∂x 2 ∂x

1 00 ∂ † 1 00 ∂
(255) σ = √ x + 00 , σ = √ x − 00 .
2 ∂x 2 ∂x
Introducing polar coordinates r, φ in the (x0 , x00 ) space
(256) x0 = r cos φ , x00 = r sin φ , 0 ≤ r < ∞, 0 ≤ φ < ∞,
and making the change of variable
(257) r 2 = (N + 1)e−ξ ,
the differential form of Morse Hamiltonian can be obtained starting from the algebraic form
Ĥ = E0 + a1 C2 (O(2)), where C2 (O(2)) = (τ † σ + σ † τ )2 , and E0 collects all terms related to the
totally simmetric operator N̂ 1. The two equations
(258) N̂ |N, M i = N |N, M i ,
(259) (τ σ + σ τ )2 |N, M i = M 2 |N, M i ,
† †
1Notice that in both Refs [40, 41] the Eqs. (254) are defined without the imaginary unit i. According to
this definition, Eq. (259) turns into (τ † σ − σ † τ )2 |N, M i = M 2 |N, M i, but this does no longer correspond to the
Casimir operator C2 (O(2)).
96 9. APPENDIXES
must be satisfied, where N and M are the quantum numbers associated to the chain U (2)(→
N ) ⊂ O(2)(→ M ). This leads directly to the coordinate equation for the ξ-dependent factor
of the function ΨN,M (ξ, φ) = RN,M (ξ)eiM φ :
" 2 #
d2 1 M2
(260) − 2+ (N + 1) (e−2ξ − 2e−ξ ) RN,M (ξ) = − RN,M (ξ) ,
dξ 2 4
that can be identified with Morse equation upon identification of algebraic and coordinate
representation parameters.
The theory of dynamical symmetry suggests that this same procedure could be inverted to
derive the algebraic form of the Hamiltonian starting from the relevant coordinate representa-
tion, i.e. from its potential function in coordinate representation. This should work at least for
a potential that does not deviate too much from a case with dynamical symmetry. For example
a potential power expansion of the form
k
X
(261) V (x) = Vi [1 − e−α(x−x0 ) ]i ,
i=2
should correspond, within the algebraic approach, to an expansion of the type:

k
X
(262) Ĥ = E0 + Ai [C1 (O(2))]i ,
i=2
where by C1 we mean the Casimir operator of degree 1. It would be very appealing if this
was the case, as a completely algebraic formulation would take great advantage of systematic
inclusion of higher-order anharmonicity in molecular problem, and one could exploit the whole
dynamical symmetry formalism to obtain multidimensional extensions.
Unfortunately, the correspondence between Eqs. (261) and (262) is not exact. If it were
exact, any arbitrary potential expanded in power series of Morse coordinates would be solvable,
and the same conclusion could be drawn for the harmonic-oscillator case.
The derivation that leads to Eq. (260), the most suitable to establish a connection between
algebraic and Schrödinger picture, is not immediately useful for an expanded potential: the
chain of coordinate transformations following Eqs. (254) and (255) involves both the real space
coordinate r and the auxiliary variable ξ, and thus any power of C1 (O(2)) in Eq. (262) leads
in to a form in which the two variables are not exactly separable, as occurs in the special
case of the (solvable) Morse potential [60]. A correspondence between the algebraic and the
geometric (coordinate space representation) form of an operator Ô is obtainable through more
complicated procedures that involve taking the expectation values of the algebraic operator on
a coherent state: the correspondence is non-exact, however, and its validity can be assessed
only through the evaluation of classical limits holding only in a limited phase space region
[70]. Different approximate method are used, all producing classical limits equivalent to the
leading order in N . As an example, we sketch here the simplified formulation based on the
so-called intensive boson operators method [71, 72]. It consists in redefining the operators
(254) and (255) multiplying them for √1N , N = total number of bosons. For these operators,
the commutation relations are:

σ σ† 1 τ τ† 1
(263) 1/2
, 1/2 = , 1/2
, 1/2 = .
N N N N N N
As a consequence, in the large-N (classical) limit, the rescaled operators commute.
Define complex numbers

τ τ†
(264) ξ→ , ξ∗ → ,
N 1/2 N 1/2
and
σ ∗ σ†
(265) η→ , η → ,
N 1/2 N 1/2
for which the definition of N as the total number of bosons requires that
(266) |ξ|2 + |η|2 = 1 .
Because the overall phase is arbitrary, it can be chosen such that η is real and non-negative,
and thus
(267) η = (1 − |ξ|2 )1/2 .
The recipe to obtain the classical limit Hcl is the following: substitutions
(268) τ → ξN 1/2 , τ † → ξ ∗ N 1/2 ,
(269) σ → (1 − |ξ|2 )1/2 N 1/2 , σ † → (1 − |ξ|2 )N 1/2 ,
lead to a Hamiltonian that , to order N1 , is equivalent to taking the expectation value of the
algebraic Hamiltonian Ĥ in the coherent states. The classical Hamiltonian will be, in general,
of the form
(270) H̃cl = p2 f (p, x) + V (x) ,
i.e. it corresponds to a Schrödinger equation with a coordinate dependent mass term:
1
(271) ≡ f (p, x) .
2µ
It is then hard to identify the potential function with V (x). If one defines the potential as
(272) V (q) ≡ Hcl |p=0 ,
the correct evaluation through the expectation values on coherent states leads to an expression
that for solvable potential has a traditional coordinate independent mass term for the special
case of a single C2 (O(2)) Casimir term.
However for a general algebraic expansion of the form (262), the corresponding classical
Hamiltonian may have more intricated mixed potential and kinetic contribution, as well as
higher order contribution, such as p4 , leading to higher order differential equations, that are
usually reduced to canonical Hamiltonian via a local approximation [70]. Due to these diffi-
culties, a straightforward rigorous generalization to potentials different from pure Morse seems
hard to achieve within the dynamical symmetry approach.
Bibliography
[1] R. Lemus and A. Frank, Chem. Phys. 242, 25 (1999).

[2] S. Carter and N. C. Handy, Comp. Phys. Rep. 5, 115 (1986).
[3] A. Del Monte, N. Manini, L. G. Molinari and G. P. Brivio, Mol. Phys. 103, 689 (2005).
[4] G. D. Carney, L. L. Sprandel, and C. W. Kern, Adv. Chem. Phys. 37, 305 (1978).
[5] G. Simons, R. G. Parr and J. M. Finlan, J. Chem. Phys. 59, 3229 (1973).
[6] P. Jensen, Mol. Phys. 98, 1253 (2000).
[7] J. W. Ellis, Phys. Rev. 32, 906 (1928).
[8] R. Mecke, Z. Phys. 81, 313 (1933).
[9] J. Tennyson and B. T. Sutcliffe, J. Chem. Phys. 77, 4061 (1982).
[10] P. Jensen, J. Mol. Spectrosc. 128, 478 (1988).
[11] J. Tennyson, M. A. Kostin, P. Barletta, G. J. Harris, O. L. Polyansky, J. Ramanlal, and N. F. Zobov,
Comp. Phys. Comm. 163, 85 (2004).
[12] S. Carter, I. M. Mills, and N. C. Handy, J. Chem. Phys. 99, 4379 (1993).
[13] M. Carvajal, J. M. Arias, and J. Gómez-Camacho, Phys. Rev. A 59, 1852 (1999).
[14] B. Molnár, P. Földi, M. G. Benedict, and F. Bartha, Europhys. Lett. 61 (4), 445 (2003).
[15] J. F. Ogilvie, Proc. R. Soc. A 378, 287 (1981).
[16] B. T. Sutcliffe, Mol. Phys. 48, 25 (1983).
[17] A. Frank, R. Lemus, F. Pérez-Bernal, and R. Bijker J., Mol. Spectrosc. 196, 329 (1999).
[18] J. C. Light and T. Carrington Jr., Adv. Chem. Phys. 114, 263 (200).
[19] J. A. Coxon and M. Molski, Phys. Chem. Comm. 20, 1 (2001).
[20] J. F. Ogilvie, Chem. Phys. Lett. 348, 447 (2001).
[21] J. A. Coxon and M. Molski, J. Mol. Spectrosc. 223, 51 (2004).
[22] R. J. Le Roy, J. Mol. Spectrosc. 228, 92 (2004).
[23] C. Schwartz and R. J. Le Roy, J. Mol. Spectrosc. 121, 420 (1987).
[24] P. M. Morse, Phys. Rev. 34, 57 (1929).
[25] A. Abramovitz and I. Stegun, Handbook of Mathematical Functions (Dover, New York (1964).
[26] J. N. Huffaker, J. Chem. Phys. 64, 3175 (1976).
[27] J. N. Huffaker, J. Chem. Phys. 64, 4564 (1976).
[28] J. N. Huffaker, J. Mol. Spectrosc. 71, 160 (1978).
[29] J. N. Huffaker and Loc Binh, Tran J. Chem. Phys. 76, 3838 (1982).
[30] J. Makarewicz, J. Phys. B: At. Mol. Opt. Phys. 24, 383 (1990).
[31] M. L. Sage, Chem. Phys. 1978, 35, 375.
[32] V. Špirko, P. Jensen, P. R. Bunker, and A. Čejchan, J. Mol. Spectrosc. 112 183 (1985).
[33] F. Pérez-Bernal, I. Martel, J. M. Arias, and J. Gómez-Camacho, Phys. Rev. A 63, 052111 (2001).
[34] F. Pérez-Bernal, I. Martel, J. M. Arias, and J. Gómez-Camacho, Phys. Rev. A 67, 052108 (2003).
[35] R. Lemus, J. M. Arias, and J. Gómez-Camacho, J. Phys. A : Math. Gen. 37, 1805 (2004).
[36] F. Cooper, A. Khare, and U. P. Sukhatme, Phys. Rep. 251, 268 (1995).
[37] S. H. Dong, and R. Lemus, Int. J. Quant. Chem. 86, 265 (2002).
[38] R. Dutt, A. Khare, and U. P. Sukhatme, Am. J. Phys. 56, 163 (1988).
[39] L. Gendenshtein, Pis’ma Zh. Eksp. Teor. Fiz. 38, 299 (1983) (JETP Lett. 38, 356 (1983)).
[40] F. Iachello and R. D. Levine, Algebraic Theory of Molecules (Oxford University Press, 1995), pag. 161.
[41] S. Oss, Adv. Chem. Phys. Vol XCIII, 455, (1996).
[42] A. Bordoni, and N. Manini, Int. J. Quant. Chem. (in press, early online view 2006).
[43] K. Patkowski, G. Murdachaew, C.-M. Fou, and K. Szalewicz, Mol. Phys. 103, 2031 (2005).
[44] W. Kolos, and L. Wolniewicz, J. Chem. Phys. 49, 404 (1968).
[45] S. Carter, J. M. Bowman, and L. B. Harding, Spectrochim. Acta A 53, 1179 (1997).
[46] J. F. Ogilvie, J. Phys. B: At. Mol. Opt. Phys. 27, 47 (1994).
99
100 BIBLIOGRAPHY
[47] J. F. Ogilvie, J. Mol. Spectrosc. 180, 193 (1996).

[48] J. A. Coxon and M. Molski, Spectrochim. Acta A 59, 13 (2003).
[49] R. Lemus, J. Mol. Spectrosc. 225, 73 (2004).
[50] S. N. Yurchenko, M. Carvajal, P. Jensen, H. Lin, J. J. Zheng, and W. Thiel, Mol. Phys. 103, 359 (2005).
[51] C. Léonard, N. C. Handy, S. Carter, and J. M. Bowman, Spectrochim. Acta A 58, 825 (2002).
[52] S. M. Colwell, S. Carter, and N. C. Handy, Mol. Phys. 101, 523 (2003).
[53] T. Rajamäki, A. Miani, and L. Halonen, J. Chem. Phys. 118, 6358 (2003).
[54] T. Rajamäki, A. Miani, and L. Halonen, J. Chem. Phys. 118, 10929 (2003).
[55] S. Takeno and G. P. Tsironis, Phys. Lett. A 343, 274 (2005).
[56] M. S. Child and L. Halonen, Adv. Chem. Phys. 57, 1 (1984).
[57] K. Takanagi, Progr. Theor. Phys. Suppl. 25, 1 (1963).
[58] K. Takanagi, Adv. Atom. Molec. Phys. 1, 149 (1965).
[59] G. W. F. Drake, and C. S. Lin, J. Phys. B 7, 398 (1974).
[60] O. S. van Roosmalen, Algebraic Descriptions of Nuclear and Molecular Rotation-Vibration Spectra, PhD
Thesis, Univ. Gröningen (unpublished 1982).
[61] F. Iachello and R. D. Levine, J. Chem. Phys. 77 , 3046 (1982).
[62] O. S. van Roosmalen, F. Iachello and R. D. Levine, J. Chem. Phys. 79 , 2515 (1983).
[63] N. C. Handy and S. Carter, Mol. Phys. 102, 2201 (2004).
[64] R. Wyatt, J. Chem. Phys. 109, 10732 (1998).
[65] J. Pochert, M. Quack, J. Stohner, and M. Willeke, J. Chem. Phys. 113, 2719 (2000).
[66] A. Callegari, R. Pearman, S. Choi, P. Engels, H. Srivastava, M. Gruebele, K. K. Lehmann, and G. Scoles,
Mol. Phys. 101, 551 (2003).
[67] M. Gruebele, J. Chem. Phys. 104, 2453 (1996).
[68] A. A. Stuchebrukhov and R. A. Marcus, J. Chem. Phys. 98, 6044 (1993).
[69] G. Pöschl, and S. Teller, Z. Physik 83, 143 (1933).
[70] M. W. N. Ibrahim, arXiv:physics/0005037 (2000).
[71] R. Gilmore, J. Math. Phys. 20, 891 (1979).
[72] R. Gilmore, Catastrophe Theory for Scientists and Engineers (Wiley, New York 1981).

Bordoni PH D

Uploaded by

Copyright:

Available Formats

Bordoni PH D

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bordoni PH D

Uploaded by

Copyright:

Available Formats

Università degli Studi di Milano

Facoltà di Scienze Matematiche, Fisiche e Naturali

Systematic calculation of molecular vibrational

Tesi di Dottorato di Andrea Bordoni PhD Thesis of

Coordinatore Gianpaolo Bellini Coordinator

Dottorato di Ricerca XIX ciclo, anno 2005-2006

Energy / depth of the well

Figure 1.1. 1D-molecular potential, and a pictorial view of a Taylor potential

• the potential expansion in normal-mode coordinates is intrinsically local: it works well

Figure 1.2. Approximations of the LJ potential (solid): the Morse potential

1.1. A simple example

n LJ Taylor 4 Taylor 4 Taylor 10 Pure Morse

The Morse oscillator: bases and matrix elements

2.1. The Schödinger equation for the Morse potential

(7) VM (r) = V0 e−2α(r−r0 ) − 2V0 e−α(r−r0 ) .

PSfrag replacements V0 (1 − e−1 )2

provide a meaningful approximation of V (x) as it approaches dissociation. As the matrix ele-

2.3. Expansion in the Morse basis

(33) Vd (x) = ai [v(x)]i ,

which for the complete potential implies the expanded form

(34) V (x) = ai [v(x)]i , where a2 ≡ V0 .

state Finite Differences Morse basis

state Finite Differences Morse basis

state Finite Differences Morse basis

k R = 0.01 R = 0.02 R = 0.20 R = 0.50

state Finite Differences Morse basis

state Finite Differences Morse basis

state Finite Differences Morse basis

2.4. Beyond the Morse basis

2.5. Morse and SUSY Shape Invariant Potentials

28 2. THE MORSE OSCILLATOR: BASES AND MATRIX ELEMENTS

Â1 Â2 Ân−1

Ĥ1 Ĥ2 ··· Ĥn

(66) V̂2 (x; a1 ) = V̂1 (x; a2 ) + R(a1 ) ,

where a1 is a set of parameters, a2 = f (a1 ), and R(a1 ) does not depend on x.

The SIP condition (66) is satisfied

2.5.4. Morse factorized Hamiltonian. We follow now a slightly different derivation of

where Fn are normalization constants.

Figure 2.10. Morse ground-state wavefunction (left), 5th state wavefunction

state Morse exact QNSB diagonalization

state Morse basis Finite Differences QNSB (Ns = 30)

state Morse Finite Differences QNSB (Ns = 30)

• the QNSB preserves the algebraic structure of Morse potential problem, it is in a

(109) |hφσj |ϕk i|2

state R = 0.01 R = 0.02 R = 0.20 R = 0.50

state Finite Differences QNSB (Ns = 30)

state Finite Differences QNSB (Ns = 40)

state Finite Differences QNSB (Ns = 30)

The potential expansion

(111) Vd (x) = ai [v(x)]i .

LJ potential Morse potential

and the third derivative at the minimum

3.1. Potential fitting schemes

v(x) even powers

Figure 3.2. Morse potential (solid) and (v(x)4,6 − 1) (dot-dashed).

(123) V (x) = VM (x) + Vd (x) , Vd (x) = a22i [v(x)]2i .

to the LJ parameters are:

Eq (124), Nmax =12

Eq. (32), Nmax = 6