New Approach for the Electronic Energies of the Hydrogen

Molecular Ion
arXiv:physics/0607081v1 [physics.chem-ph] 8 Jul 2006

Tony C. Scott1 , Monique Aubert-Frécon2 , and Johannes Grotendorst3

1
Zentralinstitut für Angewandte Mathematik, Forschungszentrum Jülich GmbH, 52425 Jülich, Germany
Institut für Physikalische Chemie, RWTH Aachen, 52056 Aachen, Germany
Institut für Organische Chemie, Fachbereich Chemie, Universität Duisburg-Essen, 45117 Essen, Germany
email: scott@pc.rwth-aachen.de

2
LASIM, CNRS et Université Lyon 1, Campus de La Doua, Bâtiment Alfred Kastler, 69100 Villeurbanne Cédex, France
email: frecon@lasim.univ-lyon1.fr

3
Zentralinstitut für Angewandte Mathematik, Forschungszentrum Jülich GmbH, 52425 Jülich, Germany
email: j.grotendorst@fz-juelich.de

February 2, 2008

Abstract

Herein, we present analytical solutions for the electronic energy eigenvalues of the hydrogen molecular
ion H+2 , namely the one-electron two-fixed-center problem. These are given for the homonuclear case
for the countable infinity of discrete states when the magnetic quantum number m is zero i.e. for 2 Σ+
states. In this case, these solutions are the roots of a set of two coupled three-term recurrence relations.
The eigensolutions are obtained from an application of experimental mathematics using Computer
Algebra as its principal tool and are vindicated by numerical and algebraic demonstrations. Finally,
the mathematical nature of the eigenenergies is identified.

PACS: 31.15.-p, 31.15.Ar, 02.70.Wz, 31.50.Bc, 31-50.Df

1 Introduction

Although, there are well established software packages in the area of quantum chemistry such as GAUS-
SIAN [1], MOLPRO [2] and GAMESS [3] which allow to obtain approximate numerical solutions to a
number of fair sized molecules, the simplest molecule namely the hydrogen molecular ion, a quantum
mechanical three-body problem, still remains mathematically intractable.

In the fixed nuclei approximation, it is well known that the Schrödinger wave equation - a second order
partial differential equation (PDE) - of the problem of one electron moving in the field of two fixed nuclei
can be separated in prolate-spheroidal coordinates [4]. These coordinates allow a separation of variables
that results in two non-trivial ordinary differential equations (ODE), and hence two eigenparameters: the
energy parameter p2 , and a separation constant A related to the total orbital angular momentum and the
Runge-Lenz vector.
We note that asymptotic expansions for small or large internuclear distances R have been obtained. A
very comprehensive presentation of the energy eigenvalues for the ground state and a number of exited
states is shown in the work of Čı́žek et al. [9]. These could almost constitute analytical solutions but
the resulting series are divergent though asymptotic [10] and therefore useful only at large internuclear
distances. Another complication is that for the homonuclear case, every gerade energy Eg (wave function
symmetric under exchange of nuclei) has a counterpart ungerade solution (wave function antisymmetric
under exchange of nuclei) whose energy Eu has exactly the same 1/R expansion. This makes the calcu-
lation of exchange energy splittings ∆E = Eu − Eg very elusive to calculate at large R, although there
are specialized methods for recovering these splittings (e.g. see [11]).

Even recently, there has been examination of series in small R limited to the ground state short-range in-
teraction energy [12] but we still have no further insight into the actual mathematical nature governing the
energy eigenvalues. We also cite the work of Demkov et al. [13] but their analytical solutions correspond
to a peculiar charge ratio depending on the internuclear distance and therefore not physically useful.

Thus, complete analytical solutions of the eigenstates of H+2 , in areas of molecular interest, such as e.g.
the region near the equilibrium internuclear distance (bond length) of the ground state remain elusive.

A wide variety of numerical methods have been used to solve the H+ 2 problem in this case. For example,
Bates, Ledsham and Stewart [5] used recursion and continued fractions. Hunter and Pritchard [6] used
matrix methods and Rayleigh quotient iteration. Madsen and Peek [7] used power series and associated
Legendre expansions to set up two equations whose simultaneous solution then gave the two eigenpa-
rameters. An accurate way to obtain energies and wavefunctions for the one-electron two-center problem
is provided by the program ODKIL conceived by Aubert-Frécon et al. [14, 15] based on a method by
Killingbeck. As of the 1980s, it was possible to calculate the eigenenergies and the eigenfunctions of
the discrete states of H+
2 with a rapid FORTRAN program. Yet, complete analytical solutions have so
far remained elusive: the classical N -body problem cannot be solved in closed form for N ≥ 3 and the
quantum counterpart is even worse by virtue of being an eigenvalue problem.

The approach used here is called “experimental mathematics”, an unorthodox approach involving multi-
disciplinary activities by which to find new mathematical patterns and conjectures. The goal in this context
is to search and find mathematical structures and patterns to be re-examined with more “rigor” at a later
stage. The level of rigor is of course relative: in dealing with a difficult problem in applied Mathematics,
we cannot approach the level of rigor demanded in number theory. Nonetheless, we desire demonstrations
sufficiently convincing to the molecular physicist.

The present work will involve a combination of methods, results and procedures from different areas.
We first start with results from what is called: dimensional scaling. It has been known for some time
that the Schrödinger wave equation can be generalized to an arbitrary number of dimensions D which
can be subsequently treated as continuous variable [17, 18]. In the limit as D → 1+ , the hydrogen
molecular ion becomes the double well Dirac Delta function model which can be solved exactly [25] in
terms of the Lambert W function [19, 20]. Dimensional scaling applied to H+ 2 has been studied at length
by Hershbach’s group [21–24], in particular, by Frantz [21], Loeser and Lopez-Cabrera [22,23]. The latter
work provides even more insight into the mathematical relationship between the real H+2 at D = 3 and its
one-dimensional limit.

Next, armed with the information provided by dimensional scaling, we will return to the real three-
dimensional formulation of Aubert et al. [15]. This formulation is re-examined using a Computer Algebra
System (CAS) within the approach of experimental mathematics: patterns and results are obtained. The
CAS used is Maple because it is readily available to us but the results could also be implemented on other
systems. The resulting series expansions are verified numerically and algebraically. In particular, we will
demonstrate that our results are independent of choice of basis and basis size and consequently completely
general. The end-result will be then analytically compared with the one-dimensional result and put on a
near equal footing allowing us to find the mathematical category to which belong the eigenvalues of H+ 2.
In view of the type of solution obtained, a tentative “physical” picture is associated with the analytical
solutions. A summary with concluding remarks is made at the end.

2 Preliminaries - Dimensional Scaling

The D → 1+ version of H+
2 [17, 18] is given by the double Dirac delta function model:

1 ∂2ψ
− − q[δ(x) + λδ(x − R)]ψ = E(λ)ψ (1)
2 ∂x2
where ZA = q and ZB = λ q. The ansatz for the solution has been known since the work of Frost [26]:

ψ = Ae−d|x| + Be−d|x−R| (2)

Matching of ψ at the peaks of the Dirac delta functions positioned at x = 0, R when (λ = 1) yields:

q − d qe−dR
=0 (3)
qe−dR q − d

and the energies are thus given by:

E± = −d2± /2 where d± = q[1 ± e−d± R ] (4)

Although, the above has been known for more than half a century, it was not until Scott et al. [25] that the
solution for d± was exactly found to be:

d± = q + W (±qRe−qR )/R (5)

where ± represent respectively the symmetric or gerade solution and the anti-symmetric or ungerade solu-
tion and W is the Lambert W function satisfying W (t)eW (t) = t [19,20]. This function first introduced by
Johann Heinrich Lambert (1728-1777), a contemporary of Euler, has been “invented” and “re-invented”
at various periods in history but its ubiquitous nature was not fully realized within the last decade or so.

For example, the W function appears in Wien’s Displacement Law of Blackbody radiation. In general, it
has appeared in electrostatics, statistical mechanics, general relativity, radiative transfer, quantum chromo-
dynamics, combinatorial number theory, fuel consumption and population growth etc. . . (e.g. see ref. [27]
and references herein).

More recently, the Lambert W function has also appeared in “linear” gravity two-body problem [28] as
a solution to the Einstein Field equations with one spatial dimension and one time dimension (1 + 1).
The present work also includes a generalization of the W function. Recent work [30] shows that the W
function can be further generalized to express solutions to transcendental algebraic equations of the form:

PN (x)
exp(±c x) = (6)
QM (x)
where PN (x) and QM (x) are polynomials in x of respectively degrees N and M and c is a constant. The
standard W function applies for cases when N = 1 and M = 0 and expresses solutions for the case of
equal charges for eq. (1) or equivalently the case of equal masses for the two-body 1 + 1 linear gravity
problem. The case of unequal charges or unequal masses corresponds to cases of higher N and M values.
This form also expresses a subset of the solutions to the three-body linear gravity problem [29, 30] where
one deals with transcendental equations of the form (6) where M, N → ∞.
Some insight into the mathematical nature of the eigenenergies of H+ 2 is revealed by the fact that the
eigensolutions for the electronic energies at D → 1+ and D → ∞ actually bound the D = 3 ground
state eigenenergy of H+2 [21, 22] as shown in Figure 1. Moreover, the latter can be estimated by a linear
interpolation formula [23]:
1 2
E3 (R) ≈ E1 ( R3 ) + E∞ ( 23R ) (7)
3 3
This formula agrees with the numerically accurate eigenenergy (as given by program ODKIL or the work
of D. Frantz) to within about 2 or 3 digits for the range of R near the bond length. The result at D → ∞
involves the extrema of a Hamiltonian expression
√ [23, eq.(58)]. We re-examined this result. One has to
consider a region of R divided by Rc = 89 3. For R < Rc , the root is determined by the root of a
quartic polynomial [23] and the result for R > Rc is determined by a sixth degree polynomial. Thus, the
result at D → ∞ is algebraic. On the other hand, the result at D → 1+ is in terms of an implicit special
function, which is the Lambert W function. Given how well this interpolation formulation works, this
already suggests what is the mathematical nature of the eigenenergies of the true hydrogen molecular ion
(D = 3).

We can state this in view of the work of Frantz et al. [21] who showed that the D-dimensional problem
could be decoupled into two coupled ODEs for 2 ≤ D < ∞ and how a particular energy eigenvalue for a
given D could exactly express the solution of another eigenvalue for an excited state at a dimension D + 2
through a precise re-scaling.

3 Three-dimensional H+
2

3.1 Starting Formulation

The Schrödinger Wave Equation for H+

2 in atomic units is given by:
   
ZA ZB
∆+2 + + 2E ψ = 0 (8)
rA rB
As mentioned before, this is separable into prolate-spheroidal coordinates:
ξ = (rA + rB )/R , 1 ≤ ξ < ∞
η = (rA − rB )/R , −1 ≤ η ≤ 1
0 ≤ φ ≤ 2π
Q1 = R(ZA − ZB )
Q2 = R(ZA + ZB )
We can write the ansatz for the eigensolution:

ψ(ξ, η, φ) = Λ(ξ)M (η, φ) = Λ(ξ)G(η)e±imφ (9)

which allow us to obtain two coupled ODEs:

m2
   
∂ 2 ∂ 2 2
(1 − η ) − + p η − Q1 η − A M (η, φ) = 0 (10)
∂η ∂η 1 − η2
m2
   
∂ 2 ∂ 2 2
(ξ − 1) − 2 − p ξ + Q2 ξ + A Λ(ξ) = 0
∂ξ ∂ξ ξ −1
where A is the separation constant and the eigenenergy E is expressed as:
p2
Eelec = − 2 (11)
R2
Note that:

lim A = −ℓ(ℓ + 1) (12)

R→0
lim p2 = 0 (13)
R→0

Although the set of quantum numbers (n, ℓ, m) - the united atom quantum numbers - can be used to
identify the eigenstates, as is the case for e.g. program ODKIL, it must be emphasized that only the
magnetic quantum number m is a good quantum number (resulting from the azimuthal symmetry of H+ 2
about its internuclear axis).

We follow the treatment of Aubert et al. [15] and consider the following basis expansion for the η coordi-
nate: X
M (η, φ) = fkm Ykm (η, φ) (14)
k=m
where Ykm are the Spherical Harmonics. Injection of the above basis into the ODE governing M in η
leads to the creation of a symmetric matrix F whose determinant must vanish when p and A satisfy the
eigenvalue problem:
[F (p, A)] |f | = 0 (15)
where
2i2 − 2m2 + 2i − 1
 
Fi,i (p, A) = −i(i + 1) + p2 − A,
(2i + 3)(2i − 1)
(i + m + 1)(i − m + 1) 1/2
 
Fi+1,i (p, A) = −R Q1 ,
(2i + 1)(2i + 3)
p2 (i + m + 1)(i − m + 1)(i + m + 2)(i − m + 2) 1/2
 
Fi+2,i (p, A) = ,
(2i + 3) (2i + 1)(2i + 5)
are the non-vanishing matrix elements of F. If Q1 = ZA − ZB = 0 i.e. the homonuclear case, then the
pentadiagonal matrix F divides in even and odd tridiagonal matrices in terms of x where x = p2 with no
explicit dependence on the internuclear distance R (although this is not true for the other ODE in ξ). For
the ξ coordinate, we use a basis of Hylleraas functions, i.e. in terms of Laguerre polynomials:
X
Λ(ξ) = e−p(ξ−1) [2p(ξ − 1)]m/2 Cn−(m/2) Lm n−(m/2) [2p(ξ − 1)] (16)
n=m/2

[Y(p, A)] |C| = 0 (17)

2 −1
[Y(p, A)] = [Q(p, A)] + p m [B(p)]

where

bi,i (p) = 4p + 2i + 1,
h m m i1/2
bi,i+1 (p) = bi+1,i (p) = − (i − + 1)(i + + 1)
2 2
and
m2
 
R Q2
ri,i (p, A) = (2i + 1) − i − 1 − 2p + + i + R Q2 − p2 + A,
2p 4
h m m i1/2  R Q 
2
ri+1,i (p, A) = − (i − + 1)(i + + 1) × −i−1
2 2 2p
ri,i+1 (p, A) = ri+1,i (p, A) .
When m = 0, the matrix is tridiagonal. For m 6= 0, one has to consider the inverse of the matrix B, which
is not a band matrix.

Of course, we realize that this choice of basis is only one of several possible choices. The results obtained
are valid provided the results are independent of the size of the basis and the choice of basis.

3.2 Recurrence Relations

The following relations apply to the homonuclear case and when m = 0 in which case, the band matrices
are purely tridiagonal matrices. These are governed by recurrence relations namely (A.1) and (A.2) of
reference [15]:

det[M0 ] = 1
det[M1 ] = m1,1
det[Mk ] = mk,k det[Mk−1 ] − mk−1,k mk,k−1 det[Mk−2 ]

Thus for m = 0, we have the following:

det(Y 0 ) = 1
R
det(Y 1 ) = − 1 − 2 p + 2 R − p2 + A (18)
p
   
R 2
det(Y k+1 ) = (2 k + 1) − k − 1 − 2 p + k + 2 R − p + A det(Y k )
p
 2
2 R
−k − k det(Y k−1 )
p
For the even ℓ case, we have:

det(F e0 ) = 1
p2
det(F e1 ) = −A
3 !
p2 8 k 2 + 4 k − 1


det(F ek+1 ) = −2 k (2 k + 1) + − A det(F ek )
(4 k − 1) (4 k + 3)
p4 (2 k − 1)2 k2 det(F ek−1 )
−4 (19)
(4 k − 1)2 (4 k − 3) (4 k + 1)

Defining det(F oi ) = det(F (i)) for the odd ℓ case, we have:

det(F o0 ) = 1
3p2
det(F o1 ) = −2 + −A
5  
 
p2 2 (2k + 1)2 + 1 + 4k
det(Fok+1 ) = −(2k + 1)(2k + 2) + − A det(F ok )
(4k + 5) (4k + 1)

p4 k2 (2k + 1)2 det(F ok−1 )

−4 (20)
(4k + 1)2 (4k − 1) (4k + 3)
Note that the radial equations for the hydrogen atom are governed by two-term recurrence relations. Thus,
it suffices to find an eigenenergy such that the coefficient ak+1 of the basis of Laguerre functions is zero.
This in effect truncates the infinite series into a polynomial and consequently closed form solutions for
the eigenstates are obtained of the hydrogen atom. This is not possible for H+
2 which is governed by
three-term recurrence relations no matter what the choice of basis.

The band matrices for H+ 2 and their determinants have been injected into a computer algebra system. The
determinants det(Y i ) and det(F i ) (even or odd) for i = 1, 2, 3 . . . are multivariate polynomial-like in A
and p. The determinants det(F i ) are true polynomials in A and p2 . On the other hand, although det(Y i ) is
a polynomial in A, it has also negative powers for p and thus akin to a Laurent series (Laurent polynomial)
in p.

It is possible to eliminate one of the unknowns by obtaining a resultant of the two determinants det(Y i )
and det(F i ). If a and b are polynomials over an integral domain, where et 2 (rational) polynomial equa-
tions in 2 unknowns A and p.
n
Y m
Y
a = an (x − αi ) b = bm (x − βi )
i=1 i=1

Then
n Y
Y m
resultant(a, b, x) = am n
n bm (αi − βj )
i=1 j=1

This can be computed from the Euclidean algorithm or determinant of a Sylvester matrix and its roots will
be common to those satisfying the original set of polynomials. Since both expressions are true polynomials
in A only, the resultant must be in A. E.g. for i = 2 (i.e. 2 × 2 matrices)

resultant(det(Y i ), det(Fei ), A) =

64/1225 p8 + 512/245 p7 − 256/245 (R − 27) p6 − 128/245 (56 R − 369) p5

+ 64/245 (2911 − 1037 R + 27 R2 ) p4 + 32/245 (13580 − 9405 R + 948 R2 ) p3
− 32/245 (140 R3 − 17780 + 24010 R − 5571 R2 ) p2 − 64/7 (20 R3 − 224 R2
+ 481 R − 161) p − 4128/7 R3 + 19968/7 R2 + 16 R4 − 2880 R + 304
+ 16/7 R (28 R3 − 347 R2 + 791 R − 252)/p + 4 R2 (20 R2 − 108 R + 85)/p2
+ 8 R3 (4 R − 9)/p3 + 4 R4 /p4
i.e. a Laurent polynomial in p with coefficients in R only. When the size of the i × i increases, the size
of the resulting expression increases dramatically (expression swell). However, from a numerical point of
view, the most useful outcome comes from numerically solving the simultaneous expressions for det(Y i )
and det(F i ) since i must be sufficiently large to give a sufficiently good result near the bond length. In
Maple, this can be done using the fsolve procedure. To find the minimum energy for the ground state,
it is a matter of getting derivatives of these determinants with respect to R. Combining the latter with the
condition:
∂ET
= 0 where ET = Eelec + 1/R (21)
∂R
∂A ∂p
we get five equations in the five unknowns R, A, p, ∂R and ∂R . The result has been calculated using a
small Maple program. In atomic units, these are:

R = 1.997193319969992 . . .
Eminimum = −0.6026346191065398 . . .

Note that the electronic energy, evaluated at R = 2.0 a.u. for comparison, is as expected exactly the
reference tabulated value of Madsen and Peek [7] i.e. −0.6026342144949. An indirect way of ascertain-
ing the accuracy of electronic energies is to use these values in an adiabatic standard scheme to obtain
vibrational energies which are directly comparable to highly accurate values provided by approaches that
do not involve the separability of the electronic and nuclear motions (e.g. [31–33] and [34, 35]). This has
been done [36] and comparisons with values from the literature are displayed in table 3.2.
 Table 1: Ground State Vibrational Energies
Energies are in a.u., differences in cm−1 (1 a.u. = 219474.63 cm−1 )
a) ref. [31], b) ref. [32], c) ref. [33], d) ref. [34], e) ref. [35]

Quantum Present Differences

System Vibrational Adiabatic Literature ∆E
Number v Values Values (cm−1 )

H+
2 0 -0.597138471 -0.597139055a) -0.13
-0.597139063123b) -0.13
1 -0.587154167 -0.587155679212b) -0.33

D+
2 0 -0.598788594 -0.5987876(11)a) +0.22
-0.598788784331c) -0.04

HD+ 0 -0.597897521 -0.5978979685d) -0.10

-0.5978979686e) -0.10

In fact, given how heavy the nuclear centers are with respect to the electron, clamping the nuclear centers
is a very good approximation for the quantum three-body problem represented by H+ 2 with the following
caveat: the approximation that the nuclei are clamped fixed in space creates a symmetry under exchange
of nuclei in the homonuclear case. A different picture arises when the movement of nuclei is considered.
The mere movement of the nuclei breaks the symmetry under exchange of nuclei and thereby leads to a
localization of the states. In this case, the work of Esry and Sadeghpour is instructive [37].

However, if one stopped here, there is no pattern from an analytical point of view. E.g. setting x = p2
and examining det(F ei ) at low order in A, we have:

at i = 2:
1/35 p2 (−70 + 3 p2 ) + (6 − 6/7 p2 )A + O A2

at i = 3:
5 2 244 2 5 4
p 1848 − 126 p2 + p4 + (−120 + p )A + O A2



p −
231 11 11
at i = 4:
1
p2 −2162160 + 173316 p2 − 2772 p4 + 7 p6


1287
6356 4 28 6
+(5040 − 1032 p2 + p )A + O A2


p −
195 143

If we look at A = 0 and grab the leading coefficient p2 , we have the sequence

−70, 1848, −2162160 . . .. Not only are the coefficients increasing dramatically in size, they also alter-
nate in sign. Although the roots A, p of these determinants det(Y i ) and det(F i ) converge with increasing
i, the actual coefficients of these determinants and especially those of the resultant increase in size becom-
ing more and more cumbersome although a CAS can handle them (up to a point).
Moreover, we have made a particular choice of basis and the combined set of polynomial-like expressions
for the determinants though numerically useful could be viewed more as a computational “model” rather
than anything truly representative of the wave function. If we stop here, we see no pattern. Insight comes
from inverting the problem.

3.3 Roots of Determinants

The three-term recurrence relations for det(Y i ) or det(F i ) cannot be solved in closed form. We start with
det(F i ) because it is easier and has no explicit dependence on R. Upon careful scrutiny of eqs. (19) and
(20), the term in det(F k−1 ) has a coefficient in p4 whereas the term in det(F k ) has terms at order p2 . Let
us assume that p is small, which is indeed the case for small R. We can therefore neglect the last term in
det(F k−1 ) and the resulting two-term recurrence relation becomes trivial to solve. It is merely a matter
of compounding the multiplicative terms of the recursion:
k−1
Y (8 j 2 + 4 j − 1) p2

k
det(Fek ) ≈ (−1) 2 j(2 j + 1) + A − (22)
(4 j − 1) (4 j + 3)
j=0

Solving for A such that det(F ek ) = 0 yields:

8j 2 + 4j − 1
A = − 2j (2j + 1) + p2 + O(p4 ) (23)
(4j − 1)(4j + 3)

We can clearly identify the R → 0 limit with ℓ = 2 j. Similarly, for the odd case, we have:
k−1
Y (8j 2 + 12j + 3)p2

k
det(Fok ) ≈ (−1) 2 (j + 1)(2 j + 1) + A − (24)
(4 j + 5)(4 j + 1)
j=0

8j 2 + 12j + 3 2
A = − (2j + 1) (2j + 2) + p + O(p4 ) (25)
(4j + 1)(4j + 5)
We can clearly identify the R → 0 limit with ℓ = 2 j +1. Thus, although ℓ is only a valid quantum number
in the united atom limit, it is nonetheless feasible to use it to identify an eigenstate as an expansion for
small p (and small R).

By the implicit function theorem, det(F i ) = 0 ⇒ A = A(p2 ). Moreover, the structure of the recurrence
relations for det(F i ) and det(Y i ) namely eqs. (19), (20) and (18) tell us that all these quantities are ith
degree polynomials in A. If one can find all the values of A such that these determinants are zero, the
latter are clearly known by the fundamental theorem of algebra. If det(F i ) as a formal series in x where
x = p2 , we can use reversion of power series to obtain an analytical solution. This is the best possible
analytical result. E.g. , we consider:

x = cos(x) ⇒ x/ cos(x) = λ where λ = 1 (26)

1 3 5 5
x + x + x + ... = λ
2 24
1 3 13 5
⇒x = λ − λ + λ + ...
2 24
The reverted series of x in terms of λ can be obtained in a number of ways including Lagrange’s method [4]
and represents the best possible representation of an analytical solution to the root of eq. (26). Formally,
the infinite series in λ is a complete solution. The issue of getting numbers for e.g. λ = 1 is a matter of a
summation technique. Solutions by reversion of power series are possible via Maple’s solve command.
E.g. inverting det(F e3 ) yields:
2 2 11 94 2
x + O x3 , x + O x3 ,



1/3 x + −6 + x−
135 21 9261
39 8
x2 + O x3


−20 + x−
77 1715
where x = p2 . To first order in x (or p2 ), we recover the solutions in eq. (23) for respectively ℓ = 0, 2, 4.
The action of inverting det(F ei ) produces i solutions to order O(xi ). E.g. if we isolate the ℓ = 0 solution
obtained from det(F ei ) for i = 3, 4, 5, 6, we obtain:
2 2
x + O x3


i=3: 1/3 x +
135
2 2 4
x3 + O x4


i=4: 1/3 x + x +
135 8505
2 2 4 26
x3 − x4 + O x5


i=5: 1/3 x + x + (27)
135 8505 1913625
2 2 4 26 92
x3 − x4 − x5 + O x6


i=6: 1/3 x + x +
135 8505 1913625 37889775
· · · · ·
· · · · ·
· · · · ·
2 2 4 26 92
i → ∞ : 1/3 x + x + x3 − x4 − x5 + . . .
135 8505 1913625 37889775
What is important to note is that the coefficients are stable! Letting i → i+1 adds a term of order O(xi+1 )
to the series and yields an extra solution for ℓ = 2 (i + 1). By re-injection of this solution to within order
O(xi ) into det(Fei ) with computer algebra, one can see that det(F ei ) is satisfied, term by term to within
that same order. Conversely, the coefficients of A for a particular choice of ℓ even can be obtained from
this simple algorithm:

1. Select value of j and ℓ = 2 j and desired order N .

2. Set a0 and a1 according to eq. (23)

a0 = −ℓ (ℓ + 1)
8j 2 + 4j − 1
a1 = x
(4j − 1)(4j + 3)

3. For i = 1 to (N − 1)
Pi k i+1 . Note that a
(a) Let Atrial = k=0 ak x + ai+1 x i+1 is symbolic and not yet determined.
(b) Substitute Atrial into det(F eℓ+i+1 ).
(c) Isolate coefficient for xi+1 .
(d) Solve for ai+1 such that this coefficient is zero.

A counterpart result also holds for the odd case of ℓ i.e. for det(F ei ). This simple algorithm allows us to
yield the series solution for A for any given choice of ℓ. At the same time, the solution of this algorithm
implies that det(Fei ) = 0 is formally solved.

It must be emphasized that increasing i merely means adding basis functions. There are no singularities
between the two nuclei of H+2 , and we can expect the wave function to be not only continuous but also
continuously differentiable in that regime i.e. we expect no surprises with the basis functions as i → ∞.
As the estimates for A and p are closer and closer to the true values of the eigenparameters, the magnitude
of the coefficients fi of eq. (15) become smaller and smaller as i → ∞. In this limit, the basis set is a
valid representation of the true wave function.

The first 10 coefficients of the series for A(x) where x = p2 for ℓ = 0 are:
2 2 4 26 92
A(x) = 1/3 x + x + x3 − x4 − x5 (28)
135 8505 1913625 37889775
513988 122264 57430742
− x6 + x7 + x8
9050920003125 11636897146875 62315584221515625
26237052532 1550889714543116
− x9 − x10
1566426840576238265625 213229853673440433908203125
and our computer algebra programs allow us to generate many more such coefficients. The first three non-
vanishing terms of the Taylor series for A(p2 ) have already been published for cases of small p consistent
with small internuclear distances R [38–40]. We now claim that the present algorithm provides a means
of generating the Taylor series of A in small x where x = p2 , the result being valid as i → ∞ and thus
independent of the size of the truncated basis. Later, we will demonstrate it to be independent of the actual
choice of basis. However, the first test concerns numerical vindication.

3.4 Numerical vindication of the Series for A(p2 )

To vindicate the series, we obtain data entries of R, p and A from program ODKIL and inject the data
entries of p into the series solutions for A. We then compare the latter with the value of A obtained
from ODKIL for a given state. This is done for the ground state and a few excited states as shown in the
following tables. The results for the ground state i.e. 1s σg (n = 1, ℓ = 0, m = 0) are shown in table 2 and
those of state 2s σg (n = 2, ℓ = 0, m = 0) are shown in table 3 demonstrating that the same series of A
for a given ℓ works for more than one state. The results for the excited state 2p σu (n = 2, ℓ = 1, m = 0)
vindicate the series solution for ℓ = 1. The results for state 3d σg (n = 3, ℓ = 2, m = 0) vindicate the
series solution for A for ℓ = 2.

In all cases, we can see that the series obtained for A(p2 ) works indeed like a Taylor series, working
very well for small p. Beyond a certain value of R, the series solution rapidly degenerates. Nonetheless,
e.g. for the ground state, the series solution works well near the bond length (around R = 2 which is
underlined) and beyond. Degradation of the series becomes apparent at R = 5.

The question arises as to whether or not the series coefficients of A(p2 ) follow a pattern. We have found
none so far. The pattern of the changing signs +, − is not one of alternating series and thus this function
is unlike all the special functions known in the literature (such as e.g. [41]).

Nonetheless, there is something of a pattern for a given series when modifying the quantum number ℓ,
term by term. The first two terms a0 and a1 follow a pattern in ℓ according to e.g. (23) for even ℓ. No
such simple pattern exists for the next term a2 . However, if one solves for a2 in terms of a0 and a1 for
a high value of ℓ, say ℓ = ℓmax , one obtains a polynomial formula for a2 . If one then substitutes the
general formulae in ℓ for a0 and a1 into this polynomial expression for a2 : it will correctly generate the
coefficients a2 not only for ℓmax but for all ℓ = 0, 1, 2 . . . ℓmax . At some point, the resulting formula
will break down for a value of ℓ > ℓmax . This “triangular” relationship, - useful because one often does
calculations within for a limited range of ℓ - indicates that:

Pℓmax (ℓ)
a2 ≈ lim ,
ℓmax →∞ Qℓmax (ℓ)
 Table 2: Convergence of “Taylor series” of A(p2 )
Ground State: 1s σg (n = 1, ℓ = 0, m = 0)

ODKIL (accurate) series (20 terms)

1R p A A
0 0 0 0
0.5 .46569679 .729927345e-1 .7299273577e-1
1.0 .851993637 .249946241 .2499467374
1.5 1.18537488 .498858904 .4988725127
] 2.0 1.48501462 .811729585 .8118596153
2.5 1.7622992 1.19023518 1.190951531
3.0 2.02460685 1.64100244 1.643819599
4.0 2.52362419 2.79958876 2.822919217
5.0 3.00919486 4.37769375 4.491055954
10.0 5.47986646 20.1332932 25.05609231
20.0 10.4882244 90.0528912 -1147.477000

Table 3: Convergence of “Taylor series” of A(p2 )

State: 2s σg (n = 2, ℓ = 0, m = 0)

ODKIL (accurate) series (20 terms)

R p A A
0 0 0 0
0.5 0.241110452 0.194282436e-1 0.1942824361e-1
1.0 0.459850296 0.711543142e-1 0.7115431427e-1
2.0 0.849546791 0.248466171 0.2484661714
3.0 1.19791141 0.510154273 0.5101542740
4.0 1.51924947 0.8535318 0.8535318053
5.0 1.82176362 1.28400188 1.284001886
10 3.19930169 5.12935962 5.127249696
15 4.51129751 12.4337232 -17315.20146

which places us beyond eq. (23) (or (25)) which determine a0 and a1 only. However, this is subject of
further exploration elsewhere.

The range of the series solution can be considerably improved by modifying the recurrence relation for
det(F e) like so:

det(F e0 ) = 1 (29)
y
det(F e1 ) = −A (30)
3 !
y 8 k2 − 1 + 4 k


det(F ek+1 ) = −2 k (2 k + 1) + − A det(Fek ) (31)
(4 k − 1) (4 k + 3)
x y (2 k − 1)2 k2
−4 det(F ek−1 )
(4 k − 1)2 (4 k − 3) (4 k + 1)
where it is understood x = y = p2 but it is only x which is treated as a perturbation. This is simply a
different representation denoted A = A(x, y) but which represents the same function A(p2 ). Modifying
 Table 4: Convergence of “Taylor series” of A(p2 )
State: 2p σu (n = 2, ℓ = 1, m = 0)

ODKIL (accurate) series (20 terms)

R p A A
0 0 -2 -2
0.5 .254186316 -1.96120498 -1.961204981
1.0 .53141962 -1.83001042 -1.830010419
2.0 1.15545177 -1.18688939 -1.186889387
4.0 2.35889913 1.53846448 1.538464473
6.0 3.43970785 5.92793017 5.927930398
8.0 4.4671459 12.0646853 12.07439611
9.0 4.97308004 15.8356448 16.60977070
10 5.476774 20.0920989 58.89905749
20 10.4882239 90.0528776 .7649129703e13

Table 5: Convergence of “Taylor series” of A(p2 )

State: 3d σg (n = 3, ℓ = 2, m = 0)

ODKIL (accurate) series (20 terms)

R p A A
0 - -6 -6.
0.5 0.166934253 -5.98541087 -5.985410869
1.0 0.335547827 -5.94115241 -5.941152409
2.0 0.686698811 -5.75530105 -5.755301048
4.0 1.51188304 -4.86085811 -4.860858108
6.0 2.37168861 -3.43229937 -3.432299419
8.0 3.09069127 -2.07684281 -2.076688125
10 3.69538523 -0.874720469 2.071237971
20 6.12806789 7.31365225 5232651466.

slightly our previous algorithm, we obtain e.g. a modified series solution for A(x, y) for ℓ = 0:

y 14 yx 14 y 2 P1 (y) x2
A= − + (32)
3 15 (2 y − 63) 375 (2 y − 63)3 (2 y − 231)

28 y 3 P3 (y) x3
− + ...
121875 (2 y − 63)5 (2 y − 231)2 (2 y − 495)
where the polynomials Pk (y) of order k are given by:

P1 (y) = 94 y − 44121
P3 (y) = 166376 y 3 + 16398492 y 2 + 131745081006 y − 13685763372435

Note that if we inject y = x into the above and make a Taylor series expansion in x, we simply recover
the series solution in x = p2 obtained in eq. (27) for ℓ = 0. Since the radius of convergence is determined
by the closest singularity or branch point in the complex plane, we have

2 y − 63 = (2 p2 − 63) = 0 ⇒ p ≈ 5.6
We note that the sequence of numbers 63, 231, 495, 855,. . . which appear in the denominator have a pattern
which can be found using the gfun package [45]. This demonstrates that these numbers fit a holonomic
function and it is found that these fit the pattern:

3 (4 j + 3)(4 j + 7) (33)

We recognize it as one of the terms which appear in the recursion relations for det(Fek )
i.e. (4k − 1)(4k + 3) with k = j + 1. However, no pattern has (so far) been found for the polyno-
mials Pk (y). Nonetheless, our computer algebra routines allow us to generate this series to relatively high
order.

Next, the sum can be calculated using non-linear transformations known as the Levin or Sidi transfor-
mations. The latter involves a series transformation by which one can accelerate the convergence of
a series and even sum divergent series (e.g. see the work of [43, 44]). We take the point of view
that a Taylor or asymptotic series has all the desired “information”, getting numbers from the series
is a matter of a summation technique. These transformations are available in the Maple system as
NonlinearTransformations.

The best results for the ground state are obtained by applying a Sidi d transformation in x compounded
with y as shown in table 6. Even when the modified series behaves badly, the result from the Sidi d
transformation provides reliable numbers. The results hold up remarkably well all the way up to R = 10
and beyond. Beyond R = 10, the asymptotic series expansions as e.g. listed by Čı́žek et al. [9] are more
useful. What is important in our case, is that our series solution works so well around the bond length and
the intermediate regime.

Table 6: Convergence of Series A(x,y)

Ground State Revisited: 1s σg (n = 1, ℓ = 0, m = 0)

A
R series (12 terms) ODKIL (accurate) Sidi-d
1 0.24994624090 0.2499462409 0.2499462410
2 0.81172958404 0.8117295840 0.8117295850
3 1.6410024366 1.6410024369 1.6410024370
4 2.7995666114 2.7995887586 2.7995887590
5 3.9638237398 4.3776938960 4.3776937530
6 -4.6313683166e+03 6.4536051398 6.4536037430
8 -3.7137673759e+12 12.2262006172 12.2261746150
10 -1.5608159299e+33 20.1339450995 20.1332931780
15 1.0054600411e+15 48.8656127918 48.8223535290

3.5 Solution for A(R, p)

Although we have eliminated one of the unknowns i.e. found A(p2 ) such that the determinantal conditions
for det(F i ) are satisfied, there is still the remaining determinant det(Y i ) to address. The recurrence
relations for det(Y i ) of eq. (18) depend on the internuclear distance R and have more structure than those
of det(F ei ) of eq. (19) or det(F oi ) of eq. (20). Nonetheless, we proceed in parallel to what we did for
A(p2 ).

To start with, we ignore the term det(Y k−1 ) and solve the resulting two-term recurrence relation since all
linear recurrence relations of this type are solvable in terms of the roots of the characteristic polynomial
obtained by assuming a det(Y i ) = f i and then solving for f :
Γ((2 k p + Y+ + X/2p)) Γ((2 k p + Y+ − X)/(2p))
det(Y i ) = (−2)k (34)
Γ(−(Y− − X)/(2p)) Γ(−(Y− + X)/(2p))
where
p
X = 2 p4 − p2 + R 2 + 2 A p2
Y+ = 2 p 2 + p − R
Y− = 2 p 2 − p + R

and Γ is the Gamma function [41]. This result bears some resemblance with the outcome of solving the
eigenvalue problem for the hydrogen atom. In this case, solutions to the ODE for the radial equation in
the radius r can be expressed in terms of hypergeometric functions. Matching the asymptotic solution at
r → ∞ with the regular solution at r → 0 necessitates the elimination of the irregular solution by forcing
one of its coefficients - also expressed in terms of the Gamma function - to be zero (e.g. see [42]). In our
case (as in the case of the hydrogen atom), it is a matter of ensuring that the arguments for one (or both)
of the Gamma functions in the denominator of the expression above to be − where  = 0, 1, 2 . . . Thus,
solving for A, we find that:
R (1 + 2 )
A(R, p) ≈ p2 + 2 (1 + 2 ) p + 1 − 2 R + 2  + 2 2 − . (35)
p
What remains is the identification of j. Next we treat term det(Y k−1 ) as a perturbation formally by
multiplying it by λ with the understanding that (λ = 1). For  = 0, the series solution for A(R, p) is:

R (p − R)2 λ
A(R, p) = (p + 1)2 − 2 R − − (36)
p 2 p (2 p2 + 2 p − R)
(p − R)2 P4 (R, p)λ2
−
(8 p (2 p + 2 p2 − R)3 (3 p + 2 p2 − R))
(p − R)2 P10 (R, p)λ3
−
(16 p (2p + 2p2 − R)5 (3p + 2p2 − R)2 (4p + 2p2 − R))
where

P4 (R, p) = 14 p4 + (13 − 12R) p3 + R (2 R − 17) p2 + 7 pR2 − R3

P10 (R, p) = 584 p10 − 8 (116 R − 233) p9 + 2 (256 R2 + 969 − 1878 R) p8
− 4 (28 R3 − 722 R2 + 1184 R − 165) p7
+ R (8 R3 + 4678 R − 1056 R2 − 1897) p6
+2 R2 (92 R2 − 1200 R + 1145) p5
− R3 (12 R2 − 678 R + 1513) p4 − 2 R4 (50 R − 297) p3
+ R5 (6 R − 139) p2 + 18 R6 p − R7

The series looks complicated and the presence of singularities at every −R+i p+2 p2 = 0 for i = 2, 3, . . .
already tell us that this function is unlike most special functions in the literature. However, the series gives
very good results as shown in table 7 with only 4 terms. It does not need any convergence acceleration
summation methods at large R. The results of Table 8 for state 2s σg (n = 2, ℓ = 0, m = 0) show us that
A(R, p) works well for large R but diverges for small R. Also shown in the table are the results of the
Sidi d transformation which considerably improves the series solution for small R.

What remains is to identify the meaning of the number j. By checking the solution for excited states, we
find out empirically that:
 = n − ℓ − 1 (37)
where n is the united atom quantum number. This number  is a valid quantum number for the separated
atom limit [8, eq.24,p.666]. Thus, just as we match the outward and inward radial solutions for the radial
ODE for the hydrogen atom by which to determine the eigenvalue, the eigensolution for H+ 2 results from
matching A(p2 ) governed by the united atom quantum number ℓ with A(R, p) governed by the separated
atom quantum number  = n − ℓ − 1.

Table 7: Convergence of A(R, p)

Ground State: 1s σg (n = 1, ℓ = 0, m = 0)

ODKIL (accurate) series (4 terms)

R p A A
0.5 0.46569679 0.729927345e-1 0.7299778055e-1
1.0 0.851993637 0.249946241 0.2499480309
2.0 1.48501462 0.811729585 0.8117297560
5.0 3.00919486 4.37769375 4.377693772
10 5.47986646 20.1332932 20.13329314
20 10.4882244 90.0528912 90.05289034
30 15.4919739 210.034597 210.0345960
40 20.4939187 380.025707 380.0257060
50 25.4951064 600.020452 600.0204512

Table 8: Convergence of A(R, p)State: 2s σg (n = 2, ℓ = 0, m = 0)

ODKIL (accurate) series (5 terms) Sidi-d

R p A A A
0 0 0 - -
0.5 0.241110452 0.0194282436 -7.5860659805e+02 0.0211395500
1.0 0.459850296 0.0711543142 -2.1405738045 0.0718499907
2.0 0.849546791 0.2484661710 0.23871213039 0.2485999772
3.0 1.197911410 0.5101542730 0.50971077859 0.5101743643
4.0 1.519249470 0.853531800 0.85348392428 0.8535343888
5.0 1.821763620 1.28400188 1.2839939077 1.2840020996
10. 3.199301690 5.12935962 5.1293596329 5.1293596444
15. 4.511297510 12.4337232 12.433723259 12.4337232589
20. 5.805158110 23.1467952 23.146795143 23.1467951431
30. 8.359177000 54.1918175 54.191817437 54.1918174372
40. 10.8899708000 97.83692290 97.836923003 97.8369230031

As suggested by Table 8 the series behaves well for large p, it is found that A(R, p) yields a stable series
in powers of 1/p. To within O(1/p7 ), the expansion for (36) is:

(4 R + 1) (2 R + 1) (16 R2 + 40 R + 23)
A(R, p) = (p + 1)2 − 2 R − + −
4p 4 p2 64 p3
(32 R2 + 68 R + 41) (64 R3 + 576 R2 + 1108 R + 681)
+ −
64 p4 512 p5
3 2
(256 R + 1432 R + 2566 R + 1593)
+ (38)
512 p6
(1280 R4 + 28160 R3 + 123680 R2 + 210448 R + 131707)
−
16384 p7
(8192 R4 + 95040 R3 + 358368 R2 + 587512 R + 371061)
+
16384 p8
(7168 R5 + 313600 R4 + 2607232 R3 + 8854496 R2 + 14149364 R + 9039151)
−
131072 p9
(65536 R5 + 1366016 R4 + 9200576 R3 + 29011472 R2 45621790 R + 29559559)
+ .
131072p10
The coefficients up to O(p−6 ) have been previously published [40] but our computer algebra programs
allow us to go much further.

3.6 Other Bases - Algebraic Vindication

Although our previous results are apparently independent of the size of the chosen basis, we must con-
sider other bases. For the η coordinate, we consider the Baber-Hassé and the Wilson bases [14] which are
described as follows.

Baber-Hassé: X
M (η, φ) = eimφ e−qη aℓ Pℓm (η) (39)
ℓ=m
The recurrence relation is given by:
(ℓ + m + 1) (ℓ − m)
[2p(ℓ + 1) + Q1 ]aℓ+1 + α1 (k)aℓ + (Q1 − 2pℓ)aℓ−1 = 0 (40)
(2ℓ + 3) (2ℓ − 1)
where for m = 0:

α1 (k) = A − p2 + ℓ(ℓ + 1)
a−1 = 0 .

Wilson: X
M (η, φ) = eimφ eqη (1 − η)m/2 (−1)k ck (1 − η)k (41)
k=0
The recurrence relation is:

2(k + 1) (k + m + 1) ck+1 + σ1 (k) ck + 2[Q1 + p (k + m) ck−1 = 0 (42)

where

σ1 (k) = A − p2 + (m + 1) (m + 2p) + k (k + 2 m + 4 p + 1)
c−1 = 0 .
Both of these bases have been implemented into the Maple system. If we consider ℓ = 0, the coefficient
aN of Baber-Hassé basis is of order O(1/pN ) and the coefficient cN of Wilson basis is of order O(p0 ).
However, if we inject our series solution for A(p2 ) into the series coefficients of both bases, we find that
both aN and cN are formally zero to within order O(pN ). This can be seen through a number of computer
algebra demonstrations. Thus our series solution for A(p2 ) also formally satisfies the recurrence relations
of these other bases, order by order in p.

For the ξ coordinate, apart from the used Hylleraas basis, there is also the Jaffé basis.

Jaffé:
 k
2 m/2 −m−1+Q2 /2p −pξ
X ξ −1
Λ(ξ) = (ξ − 1) (ξ + 1) e Dk (43)
ξ +1
k=0
The recurrence relation is:
Q2 Q2
(k + 1) (k + m + 1) Dk+1 + γ1 (k) Dk + (k − )(k + m − )Dk−1 (44)
2p 2p
where
Q2
γ1 (k) = A − p2 + Q2 − (m + 1) (2p + 1 − )
2p
Q2
− 2k (k + m + 2p + 1 − )
2p
D−1 = 0 .

Similarly, it can be algebraically demonstrated that e.g. for  = n − ℓ − 1 = 0, the 1/p series expansion
of A(R, p) formally satisfies the coefficients of the Jaffé basis for negative powers of p just as they satisfy
the Hylleraas basis. This demonstration allows us to consider another basis of importance for the η coor-
dinate, namely the Power basis:

Power:
X
M (η, φ) = eimφ e−q(1+η) (1 − η)m/2 dk M(−(k + δk), m + 1, 2p(1 + η)) (45)
k=0

M is the confluent hyergeometric function. The recurrence relation is:

Q1
(k + δk + 1) (k + δk + 1 − ) dk+1 + χ1 (k) dk (46)
2p
Q1
+ (k + δk + m)(k + δk + m − ) dk−1
2p
where
Q1
χ1 (k) = A − p2 − Q1 + (m + 1) (2p − 1 + )
2p
Q1
− 2(k + δk)(k + δk + m + 1 − 2p − )
2p
d−1 = 0 ,

and δk is an exponentially vanishing term in R and consequently we do not make the same demonstration
as for the Wilson and Baber-Hassé bases. However, when we let R → ∞ then δk → 0 and we can make
a similar demonstration as for the Jaffé basis using the 1/p expansion of A(R, p).
Granted, we have not proven this for all bases. Nonetheless, we emphasize that e.g. the Wilson basis
is very different from the Baber-Hassé basis or the Power Basis and the basis of spherical harmonics we
used as a starting point for this analysis. Moreover, the Hylleraas basis is also very different from the Jaffé
basis. These demonstrations strongly suggest basis independent results for A(p2 ) and A(R, p).

This analysis herein exploits the fundamental theorem of algebra i.e. that if one knows all the N roots
of a given polynomial say PN (x), the latter is completely defined within a scaling factor namely the
coefficient of its highest power in x. The three-term recurrence relations of eqs. (19), (20) and (18) have
a linear dependence on A for the term in dk but no dependence of A for the third term in dk−1 . Thus,
det(F ei ), det(Foi ) and det(Y i ) are ith degree polynomials in A regardless of whether or not the third
term in dk−1 is neglected. This allows us to completely account and identify the the eigenparameters of
the matrices F and Y for every discrete state.

3.7 Mathematical Classification of Solutions

So far, we have identified the functions implied by the determinants det(F i ) and det(Y i ), namely A(p2 )
and A(R, p) respectively for all discrete states where m = 0 for the homonuclear case. In view of previous
and recent work on the D → 1 version of H+ 2 and the findings in this work concerning the D = 3 version
+
of H2 , we are now equipped with the means to make the following analytical comparison. Here, we can
put the D → 1 and the D = 3 versions of H+ 2 on the same “canonical” footing:

D → 1: To reiterate the results of section 2, the energy eigenvalue is governed by an equation of the form:

d2
exp(−2 R d) = P2 (d){PN (d)} where E=− (47)
2
When the second order polynomial P2 (d) factors into a product of first order polynomials, both
sides of eq. (47) factors and the solution for d is a (standard) Lambert W function [19, 20]. When it
does not factor, the solution is a generalization of the W function reported in the work of [28]. When
the right side is a polynomial, the solution is a generalized Lambert W function [30]. The subscript
PN (d) reminds us that our generalization for the W function can accommodate a polynomial with
rational coefficients of arbitrary degree on the right side of eq. (47).
The exponential term on the left side is a reflection of the fact that outside the Dirac delta function
wells, the basis of the particle is a combination of free particle solutions which required matching
at the Dirac delta function peaks.

D=3: To summarize the results of the past few sections, the eigenparameter p, which plays an analogous
role to the parameter d of the D → 1 version of H+2 is determined from the equation:

p2
A(R, p) = A(p2 ){PN (p)} where E = −2 (48)
R2
The subscript PN (p) reminds us that we have a Taylor series for A(p2 ) with rational coefficients
which exactly matches the generalized right-hand side form of eq. (47). However, the left-side
of (48) looks very different than the left side of (47); it is the function implied by det(Y i ) and
is associated with the separated quantum number  = n − ℓ − 1. Nonetheless, like exp(−2Rd),
this function is well-defined asymptotically for large R. The right side of eq. (48) is implied from
det(F i ) for even or odd ℓ which is a united atom quantum number.
So far, the functions A(R, p) and A(p2 ) appear in the literature as expansions in terms of p−k and
pk respectively, restricted to k = 6 and for specific cases of large and small values of R [38–40]. We
can obtain series representations of both to a much greater extent in view of our computer algebra
 implementations. We have also seen that A(p2 ) can be represented as an infinite series in x where
x = p2 and is consequently polynomial-like.
Note that if m 6= 0, the governing equation has the same form as eq. (48) but the left side is more
complicated and more difficult to get, as the determinant det(Y i ) is no longer governed by a simple
recurrence relation. However, in principle, eq. (48) governs the entire homonuclear case.

Mathematically, in both cases, the right side of the governing equation is expressed in terms of only one of
the eigenparameters whereas the left side requires the parameter and the value of the internuclear distance
R which is determined on input.

We therefore come to the conclusion that the eigenparameter p, like its D → 1 counterpart d, is also deter-
mined by a special function which is an implicit function, an even greater generalization of the Lambert W
function. So far, the functions A(R, p) and A(p2 ) do not appear in the literature. However, we can obtain
series representations of both to the extent of getting reliable numbers, as demonstrated by our tables of
values.

On the subject of implicit functions or implicit equations, these are seen in a number of specific contexts:

Retardation Effects: Equations of form e−c λ = P2 (λ) and more generally e−cλ = P2 (λ)/Q1 (λ) ex-
press the solutions of a huge class of delayed differential equations [46, eq.(3)].

Bondi’s K-calculus: It is well known in the area of special relativity that the Lorentz transformation can
be derived from the theory of implicit functions with minimal assumptions of continuity [47]. Here
one seeks the function f (t) satisfying f (f (t)) = k2 t and the requirements that it be monotone
increasing and continuous. The unique solution is:

f (f (t)) = k2 t where v/c = k2 − 1/k2 → f (t) = k t

GRT/QFT: As we mentioned before, the Lambert W function and its generalization appear in General
Relativity as solutions to respectively the two-body and three-body linear (1 + 1) gravity problems
via dilaton theory [28, 29].

Implicit functions often appear in problems with retardation effects, relativistic or otherwise. Thus, with
some reservations, we associate with this mathematical category a tentative “physical picture”:

Although the hydrogen molecular ion H+ 2 in the context of the Schrödinger wave equation is not a rela-
tivistic formulation, the eigensolutions we obtain nonetheless suggests something akin to a retardation or
delay effect. This is not the case for a one center problem like the hydrogen atom but this characteristic
appears for a two-center problem. However, this statement must be tempered with the fact that e.g. the
Lambert W function also appears in many other types of problems with no relationship to retardation
effects.

4 Summary/Conclusions

Through experimental Mathematics using computer algebra as a tool, we have identified the mathemati-
cal structures governing the energy spectrum of the hydrogen molecular ion H+
2 for the two-center one-
electron problem.

In the present work, we started with a particular choice of basis and expressed the determinantal conditions
by which the eigenparameters p and A are obtained. From one of the two determinants, we inverted the
problem to obtain a series representation of the separation constant A(p2 ) associated with the united atom
quantum number ℓ. We applied a similar approach to obtain A(R, p) from the remaining determinant and
associated with the separated atom quantum number  = n − ℓ − 1. where n is a united quantum number.
We then demonstrated that the results were independent of the size and even the choice of basis.

The eigenparameter p for which E = −2p2 /R2 is obtained by matching A(p2 ) = A(R, p) and found
to be the solution of an implicit function, with features similar to that of the Lambert W function and its
recent generalizations [30]. This allowed us to mathematically categorize the eigenvalues (or rather make
us realize what they are not) and even to associate a tentative “physical picture” to the solutions. While
we made no pretense at rigor, the solutions were nonetheless vindicated numerically and by algebraic
demonstrations with computer algebra.

The results express analytical solutions for the ground state and the countable infinity of discrete states of
H+2 for the homonuclear case when the magnetic quantum number m = 0. From the discussion below
eq. (48), we anticipate that the eigensolutions for m 6= 0 for the homonuclear case to be qualitatively
similar though admittedly this remains to be proven. We emphasize that although the basis and approach
used here were ideal for m = 0 and the homonuclear case. the computer algebra methods shown are
directly applicated to the heteronuclear case with m 6= 0. For m 6= 0, one should work directly from the
recurrence relations of the chosen basis now that we understand how these basis coefficients behave with
better and better accuracy for the series expansions of A(p2 ) in p and the asymptotic series expansions for
A(R, p) in 1/p.

However, we make no pronouncements concerning the nature or mathematical category of the solutions
for the heteronuclear case or when the nuclei are allowed to move. We note that for m = 0, the matrix
for det(Y i ) remains tridiagonal while the band matrix for det(F i ) is pentadiagonal and consequently
governed by nested recurrence relations [15] suggesting that the analysis shown herein is possible.

A number of issues arise from this result. In a sense, the result is both overdue and premature. It is
overdue because of our present capacity to find solutions to fair-sized molecules using computational
chemistry. On the other hand, it is premature. The functions we found A(p2 ) and A(R, p) do not seem
to resemble anything we have seen in the literature. The apparent singularities or “resonances” at 2p2 =
3 (4 j + 3)(4 j + 7) for A(p2 ) and R = (j + 2) p + 2 p2 for A(R, p) for j = 0, 1, 2, . . . do not constitute
a problem since the eigenparameters p and A for a given R are never found on these resonances. Once
a value of R is injected into A(p2 ) = A(R, p), solving for p numerically did not create any problems
in the test cases examined so far. At any rate, the tables shown herein merely illustrate the convergence
properties of the functions A(p2 ) and A(R, p) we have identified: solving the coupled set of polynomials
det(F i ) and det(Y i ) for p and A at a given distance R involves no resonances and is still the most useful
method from a computational point of view. In principle, the latter can go further than any FORTRAN
program.

We have ordered series representations to relatively high order of both of these functions A(p2 ) and
A(R, p) and we can generate reliable numbers for a number of discrete quantum states. We have also
demonstrated that we could use these series beyond their radius of convergence using techniques for han-
dling divergent series.

From here, one could explore and seek alternate representations of these functions with better convergence
properties especially at low R for A(R, p) and large R for A(p2 ) but the results from the Sidi transfor-
mations are already very promising. At any rate, the hydrogen molecular ion for clamped nuclei can be
entirely contained within simple computer algebra sessions, not much more complicated than those of the
hydrogen atom1 .

The exploratory and roundabout way by which we found our solutions, suggests there is something miss-
1
Maple CAS programs are available upon request.
ing in the mathematical physics or the methods for obtaining the eigenvalues of the Schrödinger wave
equation. There is hardly any existing “technology” for solving quantum chemistry problems involving
implicit functions. Our use of a basis is certainly valid to demonstrate or prove a result. Furthermore,
the convergence of the bases used here has been confirmed by determining the asymptotic behavior of
the expansion coefficients of the wavefunctions for the various basis sets considered [40]. Nonetheless, a
more direct way of generating the functions of A(p2 ) and A(R, p) would be instructive.

Acknowledgments

One of us (T.C.S.) would like to thank Professor Arne Lüchow of the Institut für Physikalishe Chemie,
RWTH Aachen and Professor Georg Jansen of the Institut für Organische Chemie of the University of Es-
sen for their wonderful hospitality and support for allowing this work to be possible. Special thanks go to
Marc Rybowicz of the University of Limoges for helpful discussions concerning the recurrence relations
and Bruno Salvy for his “gfun” package and discussions with Philippe Flajolet on asymptotic expansions.
We would also like to thank Dirk Andrae of the Theoretical Chemistry group at the University of Biele-
feld (Faculty of Chemistry), and Greg Fee of the Center of Experimental and Constructive Mathematics
(CECM) of Simon Fraser University, for helpful information.

References

[1] http://www.gaussian.com

[2] Molpro quantum chemistry package

http://www.molpro.net

[3] The General Atomic and Molecular Electronic Structure System (GAMESS)
http://www.msg.ameslab.gov/GAMESS/GAMESS.html

[4] G. B. Arfken, Mathematical Methods for Physicists, 2nd ed., Academic Press, New York (1970).

[5] D. R. Bates, K. Ledsham and A. L. Stewart, Phil. Trans. Roy. Soc. A246, p. 215 (1953).

[6] G. Hunter and H. O. Pritchard, J. Chem. Phys. 46, p. 2146 (1967).

[7] M. M. Madsen and J. M. Peek, Atomic Data, 2, p. 171 (1971).

[8] J. D. Power, Phil. Trans. R. Soc., 274, 663 (1973).

[9] J. Čı́žek, R. J. Damburg, S. Graffi, V. Grecchi, E. M. Harrel II, J. G. Harris, S. Nakai, J. Paldus, R.
Kh. Propin, and H. J. Silverstone, Phys. Rev. A 33, pp. 12-54 (1986).

[10] J. D. Morgan III and B. Simon, Intern. J. Quantum Chem. 17, pp. 1143-1166 (1980).

[11] T. C. Scott, A. Dalgarno, and J. D. Morgan III, Phys. Rev. Lett. 67, pp. 1419-1422 (1991).

[12] M. Battezzati and V. Magnasco, J. Phys. A. 37 pp. 10639-10651, (2004).

[13] Yu. N. Demkov and I. V. Komarov, Teoreticheskaya i Matematicheskaya Fizika (USSR) (trans.
Theoretical and Mathematical Physics) 38, no.2, pp. 174-176, (1979).

[14] G. Hadinger, M. Aubert-Frécon and G. Hadinger, J. Phys. B. At. Mol. Opt. Phys. 22, pp. 697-712
(1989);
[15] M. Aubert, N. Bessis and G. Bessis, Phys. Rev. A. 10, pp. 51-58 (1974).

[16] K. Goedel, Anzeiger der Akad. d. Wiss. in Wien (math.-naturw. Kl.), no. 19 (1930).
http://home.ddc.net/ygg/etext/godel/godel3.htm

[17] D. R. Herrick and F. H. Stillinger, Phys. Rev. A 11, 42 (1975).

[18] W. N. Whitton and W. Byers Brown, Int. J. Quantum Chem. 10, pp. 71-86 (1976).

[19] R. Corless, G. Gonnet, D. E. G. Hare and D. Jeffrey, MapleTech 9, (Spring 1993).

[20] R. Corless, G. Gonnet, D. E. G. Hare, D. Jeffrey and D. Knuth, Advances in Computational Mathe-
matics, 5, pp. 329-359 (1996).

[21] D. D. Frantz and D. R. Hershbach, J. Chem. Phys. 92, pp. 6668-6686 (1990).

[22] D. D. Frantz, D. R. Hershbach and J. D. Morgan III, Phys. Rev. A 40, p. 1175 (1989).

[23] M. López-Cabrera, A. L. Tan and J. G. Loeser, J. Phys. Chem. 97, pp. 2467-2478 (1993).

[24] M. López-Cabrera, D. Z. Goodson, D. R. Hershbach and J. D. Morgan III, Phys. Rev. Lett, 68, pp.
1992-1995, (1992).

[25] T. C. Scott, J. F. Babb, A. Dalgarno, and J. D. Morgan III, J. Chem. Phys. 99, pp. 2841-2854 (1993).

[26] A. A. Frost, J. Chem. Phys. 25, 1150 (1956).

[27] S. R. Cranmer, Am. J. Phys. 72 pp. 1397-1403 (2004).

http://cfa-www.harvard.edu/˜scranmer/News2004/

[28] R. Mann and T. Ohta, Phys. Rev. D. 55, pp. 4723-4747 (1997).

[29] J. J. Malecki and R. B. Mann, Phys. Rev. E. 69, pp. 1-26, (2004).
http://arxiv.org/pdf/gr-qc/0306046

[30] T. C. Scott, R. B. Mann and R. E. Martinez, General Relativity and Quantum Mechanics: Towards
a Generalization of the Lambert W Function, AAECC, 16, no. 6, (2006).

[31] J. Shertzer and C. Green, Phys. Rev. A. 58, pp. 1082-1086 (1998).

[32] B. Gremaud, D. Delande and N. Billy, J. Phys. B. 31, p. 383 (1998).

[33] R. E. Moss, J. Chem. Soc. Faraday Trans. 89, p. 3851 (1993).

[34] S. Bubin, E. Bednarz, L. Adamowicz, J. Chem. Phys. 122, p. 041102 (2005).

[35] L. Hilico, N. Billy, B. Gremaud, D. Delande, Eur. Phys. J. D 12, p. 449 (2000).

[36] A. Yanacopoulo, Thesis no. 63.91, University of Lyon 1, France (1991).

[37] B. D. Esry and H. R. Sadeghpour, Phys. Rev. A 60, pp. 3604-3617 (1999).

[38] D. I. Abramov, S. Yu. Slavyanov, J. Phys. B. 11, p.2229 (1978).

[39] W. G. Baber, H. R. Hassé, Proc. Camb. Soc. 31, p. 564 (1935).

[40] G. Hadinger, M. Aubert-Frécon, G. Hadinger, Eur. Phys. J. D 4, p. 63 (1998).

[41] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions (9th printing), Dover, New
York (1970).
[42] H. Buchholz, Die Konfluente Hypergeometrische Funktion, Berlin, Springer (1953).

[43] J. Grotendorst, Comput. Phys. Commun. 67, pp. 325-342 (1991).

[44] E. J. Weniger, Comput. Phys. Rep. 10, pp. 189-371 (1989).

[45] B. Salvy and P. Zimmermann, Gfun: a Maple package for the manipulation of generating and
holonomic functions in one variable, ACM Transactions on Mathematical Software, 20, no. 2, pp.
163-177, (1994).

[46] S. A. Campbell, Dynamics of Continuous Discrete and Impulsive Systems, 5, pp. 225-235 (1999).

[47] H. Bondi, Relativity and Common Sense: A New Approach to Einstein, Dover, New York, (1980);
http://www.geocities.com/ResearchTriangle/System/8956/Bondi/intro.htm
 Energy vs. R for H+
2 (a.u.)

D →∞ D →1 D=3

-0.6

-0.8

-1

-1.2
E

-1.4

-1.6

-1.8

-2
0 2 4 6 8 10
R
1 2
E3 (R) ≈ E1 ( R3 ) + E∞ ( 23R )
3 3
Reference: Lopez-Cabrera, Tan and Loeser, J. Phys. Chem. 97, 2467-2478 (1993).