a

r
X
i
v
:
1
1
0
6
.
0
6
3
7
v
1


[
m
a
t
h
-
p
h
]


3

J
u
n

2
0
1
1
1
Quantum mechanical potentials exactly solvable
in terms of higher hypergeometric functions. I:
The third-order case
S. Trachanas
Foundation of Research and Technology Hellas (FORTH)
and Department of Physics, University of Crete, Crete, Greece
Abstract. We present a new six-parameter family of potentials whose solutions are
expressed in terms of the hypergeometric functions
3
F
2
,
2
F
2
and
1
F
2
. Both the
scattering data and the bound states of these potentials are explicitly computed
and the peculiar properties of the discrete spectrum are depicted in a suitable
phase diagram. Our starting point is a third-order formal eigenvalue equation of
the hypergeometric type (with a suitable solution known) which is transformed to
the Schrdinger equation by applying the reduction of order technique as the crucial
rst step. The general preconditions allowing for the reduction to Schrdinger form
of an arbitrary eigenvalue equation of higher order, are discussed at the end of the
article, and two universal features of the potentials arising this way are also stated
and discussed. In this general scheme the Natanzon potentials are the simplest
special case, those presented here the next ones, and so on for potentials arising
from equations of fourth or higher order.
I. INTRODUCTION
The search for exactly solvable quantum mechanical potentials has a long history
dating back to the early days of quantum mechanics. In 1971, Natanzon
1
appeared
to have closed the subject by explicitly constructing the full set of potentials V (x)
whose solutions can be expressed in terms of the hypergeometric functions
2
F
1
and
1
F
1
.
The subject received a new impetus by the addition to the available tools for
exact solution of the Darboux transformation
2
: A mapping of the Schrdinger equa-
tion to itself which can be applied to any solvable potential and produce an innite
chain of new ones, usually with predetermined spectral changes with respect to the
initial potential (addition or removal of bound states, etc.).
Given though that the solutions of these "derived potentials" are also expressed
in terms of the hypergeometric functions
2
F
1
and
1
F
1
or, to be precise, in terms
of linear combinations of these functions and their derivatives
3
it is reasonable
to view these potentials as simple "derivatives" of the initial ones. So, a complete
description of the now-known exactly solvable potentials is, we believe, this: Exactly
solvable are the Natanzon potentials and their Darboux derivatives.
In view of the above, it is clear that new solvable potentials can be discovered
only if we enlarge the set of functions within which their solution is sought for.
And the most natural such enlargement is the full set of hypergeometric functions
p
F
q
. These are certainly equally exact functions as
2
F
1
or
1
F
1
, the only dierence
being that they satisfy linear dierential equations of order higher than second.
Specically, the function
p
F
q
satises an equation of order q + 1 where for any
given q the index p takes the values 0 p q +1 characterizing the various types
of hypergeometric functions of the given order.
In particular, if we restrict ourselves to hypergeometric functions of the third
order e.g., the functions
3
F
2
,
2
F
2
,
1
F
2
and
0
F
2
then it is clear that their use for
constructing new exactly solvable potentials presupposes, as a necessary rst step,
the reduction of order of the third-order equation satised by these functions. But it
2
is also necessary that this initial third-order equation has an eigenvalue parameter
suitably located so that it takes the position of energy in the nal Schrdinger
equation produced by the reduction. How this can be done will be examined in the
following section.
II. THIRD-ORDER EIGENVALUE EQUATIONS REDUCIBLE TO THE SCHRDINGER
FORM.
As is well-known, if a particular solution y = y
0
of a linear dierential equation
is available, the transformation y = y
0
Y that is, the factoring out of the known
solution eliminates the Y term of the new equation and so the further substitution
Y

= u lowers its order by one. In the case of a linear equation of order three
a(x)y

+ b(x)y

+ c(x)y

+ d(x)y = 0 (1)
the application of this procedure leads to the second order equation
u

+
_
b
a
+ 3
y

0
y
0
_
u

+
_
c
a
+ 2
b
a
y

0
y
0
+ 3
y

0
y
0
_
u = 0, (2)
where u = (y/y
0
)

. With the further substitution

u(x) = y
3/2
0
exp
_

1
2
_
b
a
dx
_
U(x) (3)
equation (2) is transformed into the canonical form
U

+
_
c
a

1
2
_
b
a
_

1
4
_
b
a
_
2
+
1
2
b
a
y

0
y
0
+
3
2
_
y

0
y
0
_

+
3
4
_
y

0
y
0
_
2
_
U = 0 (4)
where the rst derivative term is missing. From (4) it is now clear that it can be
cast into the so-called Liouville form
U

+
_
w(x) v(x)
_
U = 0 (5)
which is readily reduced to the Schrdinger equation if the following conditions
are met: i) The coecients a(x), b(x) and the special solution y = y
0
of (1), are
independent of the eigenvalue parameter , ii) the coecient c depends linearly on
. That is,
c(x, ) = c
0
(x) + c
1
(x). (6)
If these conditions are met then (4) is indeed a Liouville equation with weight
function w(x) = c
1
(x)/a(x) and Liouville potential this is a suitable name for the
function v(x) in (5) as follows
v(x) =
c
0
a
+
1
2
_
b
a
_

+
1
4
_
b
a
_
2

1
2
b
a
y

0
y
0

3
2
_
y

0
y
0
_

3
4
_
y

0
y
0
_
2
. (7)
3
As far as equation (5) is concerned it is well known that it can be reduced to the
Schrdinger form

(z) +
_
V (z)
_
(z) = 0 (8)
where primes stand now for derivatives with respect to z with the so-called
Liouville transformation
(z) =
_
x(z)
_
= w
1/4
U(x)

x=x(z)
(9)
where the nal position variable z is dened by
z

(x) =
_
w(x) z(x) =
_
_
w(x) dx (10)
and the potential V (x) from now on we will always assume that x = x(z) is given
by
V (x) =
v(x) +{z, x}
w(x)
, (11)
where {z, x} is the so-called Schwartzian derivative or simply the Schwartzian of
the function z(x) dened by
{z, x} =
1
2
_
z

1
4
_
z

_
2
=
1
4
_
w

w
_

1
16
_
w

w
_
2
. (12)
It is now clear from (12) that the integration constant in (10) is eliminated and can
be ignored. Concerning the requirement that the special solution y
0
is independent
of in spite of the fact that appears in the equation this leads also to the
condition that the coecient d in (1) is linearly dependent on like c. That is,
d(x, ) = d
0
(x) + d
1
(x), (13)
whence equation (1) can nally be written as
(L + M)y = 0, (14)
where the operator L has the full order of the equation that is, three while M is
of rst order. We claim therefore that any formal i.e., with no boundary conditions
specied eigenvalue equation of the form (14) can be reduced to the Schrdinger
equation, provided that the solution y = y
0
(x) of the rst-order equation My = 0
is also a solution of Ly = 0, and therefore also of the complete equation (14)
for all values of the parameter . A suitable generalization of this proposition for
equations of arbitrary order will be given in section VII.
III. EXACTLY SOLVABLE EIGENVALUE EQUATIONS OF THIRD ORDER AND THE AS-
SOCIATED POTENTIALS
The next step is obvious. Starting with the most general third-order linear equa-
tion, solvable in terms of hypergeometric functions, we write it in the form (14)
with its parameters restricted so that the solution of My = 0 is also a solution
of Ly = 0 and then transform it into Schrdinger form producing along the way
4
the respective solvable potential. The most general third-order equation we need is
written as
x
2
(1 + x)y

+ x( + x)y

+ ( + x)y

+
_

x
+
_
y = 0 (15)
whose form is readily recognizable if we introduce the concept of dimension d =
mn of the typical term x
m
y
(n)
(of an arbitrary linear equation with polynomial
coecients) and say that (15) is a bidimensional equation. That is, an equation
whose terms can be grouped in only two dierent sets, each set having a denite
dimension. Bidimensional equations are important because their solution in power
series leads to a two-term recursion formula e.g., a
k+
= f(k)a
k
that allows us
to determine the general series coecient a
k
in closed form as a function of k.
Thus the series represents an exactly known function and the pertinent equation
can be classied as exactly solvable on purely mathematical grounds. Furthermore,
since the dimension d of the typical term x
n
y
(n)
is but the displacement on the
exponent of a typical power x
k
brought about by the operator L = x
m

n
, it
follows that for a bidimensional equation with dimensions d
1
and d
2
, the integer
in the recursion relation a
k+
= f(k)a
k
which is clearly equal to the dierence of
the two displacements experienced by the general power of the series-solution when
substituted into the equation will be = d
2
d
1
(d
2
> d
1
by convention). The
quantity will be called the step of the equation since it indeed tells us that the
series-solution proceeds in steps of size . Based on the preceding discussion, Eq.
(15) is uniquely specied by the statement that: It is the most general bidimensional
equation of third order and step unity.
Concerning our assertion that Eq. (15) is solvable in terms of hypergeometric
functions, this follows directly from the existence of a two-term recursion relation
and will be discussed further in Appendix A where we will also provide a simple
recipe for arriving at the solution using only elementary algebraic operations. Note
also that the exponents of power behaviors at zero and innity, for any bidimen-
sional equation, are determined by the conditions
L
1
x

= L
1
()x
+d1
= 0, L
2
x

= L
2
()x
+d2
= 0 (16)
i.e., from the roots of the characteristic polynomials L
1
() and L
2
() of the uni-
dimensional operators L
1
and L
2
representing the components with the least and
the largest-dimension respectively, of the bidimensional operator L = L
1
+ L
2
.
From the preceding discussion it follows that our starting point should be the
solvable, third-order eigenvalue equation
x
2
(1 + x)y

+ x( + x)y

+ ( + x)y

+
_

x
+
_
y+

_
( + x)y

+
_

x
+
_
y
_
= 0 (17)
which has the same form as eq. (14) with
L = x
2
(1 + x)
3
+ x( + x)
2
+ ( + x) +
_

x
+
_
, (18)
and
M = ( + x) +
_

x
+
_
( = d/dx). (19)
5
where , , and are new, arbitrary parameters, independent of , , and
of Eq. (18) but with the same role as these, hence the similar notation. With
M given by the expression above, the general solution of My = 0 takes the form
y = y
0
= x
s
(x+)
s

with suitably chosen s, s

and but here we will restrict our

discussion to the case s

= 1 since then y
0
can certainly be also a solution of Ly = 0
as it is a terminating series about x = 0 and the equations of hypergeometric type
are especially these admitting such solutions.
By regarding now s and as given parameters and requesting that y
0
= x
s
(x+)
be a simultaneous solution of the equations My = 0 and Ly = 0, we nd that the
operator M can only have the form
M =
_
1 +
x

_

_
s
x
+
s + 1

_
(20)
while L has the form of (18), with the parameters , and no longer independent,
but expressed in terms of , , , and s as follows:
=
1

_
3(1 )s
2
+ (2 2 + 3 3)s
_
=
_
s
3
+ ( 3)s
2
+ (2 )s
_
=
1

_
(3 2)s
3
+ (2 )s
2
+ (2 + 2 3)s
_
.
(21)
As for the parameter it can be set equal to zero since it only appears as an
additive constant to the potential and can therefore be ignored.
Our assertion becomes now quite specic: We claim that the formal eigenvalue
equation of third order
x
2
(1+x)y

+x(+x)y

+
_
x +
_
1 +
x

__
y

+
_
s
x
+
1 + s

_
y = 0
(22)
with , and as in (21) is reduced to a Schrdinger equation which must
therefore be solvable, since Eq. (22) is solvable. More specically, the expression
w(x) = c
1
(x)/a(x) implies that the weight function in the relevant Liouville equa-
tion (Eq. (5)) will be equal to
w(x) =
1 + (x/)
x
2
(1 + x)
z(x) =
_
_
w(x) dx =
_
x0
ln x x = e
z
x
1

ln x x = e

z
(23)
where we have assumed and to be positive so that the change of variables
z = z(x) maps the region 0 < x < between the singular points x = 0 and
x = of the initial equation into the full region < z < of the nal
position variable z. As for the complete expression of the function z = z(x), this is
a simple elementary function which, neverthelles, does not seem to be invertible in
terms of elementary functions even though the existence of an inverse is guaranteed
by the fact that z

(x) > 0 in the region of positive x. But just as in the case of

Natanzon potentials, this poses no problem in solving the Schrdinger equation
and calculating its spectral properties, since only the asymptotic forms of x(z) are
involved in these calculations and these are explicitly known (Eq. (23)). We note
6
also that the initial variable x is much more suitable for working out the solution;
so we will use this variable henceforth in the understanding that at the end one
has to make the substitution x x(z).
Based on the preceding discussion, we can now readily construct the potentials
derived from (22) by simply applying formulas (1) through (12). Let us begin with
the Schwartzian term V
s
(x) = {z, x}/w(x) which is given by
V
s
(x) =

16
4
2
x
4
+ (4 + 12
2
)x
3
+ (3 + 18 + 3
2

2
)x
2
+ (12 + 4
2
)x + 4
2
(x + )
3
(1 + x)
(24)
while for the full potential V (x) =
_
v(x)/w(x)
_
+ V
s
(x) we get
V (x) =
Ax
4
+ Bx
3
+ Cx
2
+ Dx + E
(x + )
3
(1 + x)
(25)
whereby
A =
_
9
4
3
_
s
2
+
_
3
2
2 3 + 3
_
s +
_

4
+
_

B =
_
9
2

9
2
3
_
s
2
+
+
_

9
2

2
+ 3 + 3
2
+
3
2

9
2
2 + 3
_
s
1
2
_

2
3
2
+ 3 + 3
_
+
1
4
(3 + 1)
C =

4
_
9(
2

2
3)s
2
+ 6(4 3 2 + 2 +
2

2

2
)s
+
2

2
2
2
+ 4 2 10 +
2
4
+
3
4

2
+
9
2
+
15
4
_
D =

2
4
_
6( 3)s
2
+ (6 2 12 + 18)s
+(2 4 + 2
2
6 + + 3)
_
E =
_

3
4
s
2

1
2
( 3)s +
1
4
( 1)
2
_

3
.
(26)
Note that even though the potential is a simple rational function of the initial
variable x, it nevertheless depends in a complicated way on the ve dimensionless
parameters , , , and s of the problem at hand. (There are ve and not six
parameters as we had initially claimed, since the sixth is a scaling parameter of
z which we implicitly set equal to unity see, e.g., the rst numerical coecient
in (20) so that, together with the substitutions = 2m = 1, we can arrive at a
complete system of units for the Schrdinger equation.) The form of V (z) which
is but the form of V (x) in the interval 0 < x < stretched to < z < +
7
will typically be as shown in Figure 1 with V
0
(= E/
3
) and V

(= A/) given by
V
0
V

=
3
4
s
2

1
2
( 3)s +
1
4
( 1)
2
V

=
_
9
4
3
_
s
2
+
_
3
2
2 3 + 3
_
s +

2

4
+ .
(27)
Even though the expression for the potential in terms of its parameters is quite
complex, the solutions of the respective Schrdinger equation look much simpler!
To a large extent, this simplication is due to a dierent parametrization of the
problem which occurs naturally once we realize that Eq. (22) has x
s
as a solution
provided takes the value

0
=
_
3s
2
+ (2 3)s
_
(28)
V(z)
V
0
V

V

z
Figure 1: A typical form of the potential V (z)
so that the corresponding eigenfunction
0
(x) is

0
(x) =
x
q
(1 + x)
rq+(1/4)
(x + )
1/4
, (29)
where q and r are given by the expressions
q =
3s + 1
2
, r =
3s + (/) 2
2
. (30)
But since for x 0 and x we have
0
(x) x
q
and
0
(x) x
r
respectively,
it follows that (29) will satisfy the boundary conditions at the origin and at innity
i.e., at and + of the variable z only if q > 0 and r < 0. In this case,

0
(x) represents the systems ground state since it has no nodes in this region. The
crucial nding now is this: If we take the value (28) as the reference level for the
potential energies and eigenvalues of the problem, using instead of V and the
quantities
U = V
0
, =
0
(31)
then a remarkable simplication ensues. U and do not depend separately on the
parameters s, and , but only on their combinations q and r. In other words, when
8
we make this (parametrically dependent) change of reference level, the number of
(dimensionless) parameters of the problem gets reduced from ve to four ; q, r,
and . Note, for example, that subtracting
0
from the asymptotic values (27)
gives
U
0
V
0

0
= q
2
, U

0
= r
2
, (32)
which are much simpler expressions than those before, with the parameters now
being q, r, and . Lets call this way of parametrizing the problem invariant
parametrization; we shall work with this from now on. The expression for the po-
tential U(x) now becomes as in (25), i.e.,
U(x) =
Ax
4
+Bx
3
+Cx
2
+Dx +E
(x + )
3
(1 + x)
(33)
but with new coecients A, B, C etc., which are given by
A =
2
r
2
B = 2
2

2
r
2
+ 2qr q +
2

2
r +
1
4
(1 )
C =
_
q
2
+
2

2
r
2
+ 4qr (1 + )q + (1 + )r +
3
16
(1 )
2
_
D =
2
_
2q
2
+ 2qr q + r +
1
4
( 1)
_
E =
3
q
2
.
(34)
.
IV. RESULTS: THE FULL HYPERGEOMETRIC CASE
As explained in Appendix A, if we use the above parametrization, the solution
(x) of the Schrdinger equation satisfying the boundary condition (x = 0)
(z =) =0 in the discrete spectrum region or (z ) e
ikz
in the case
of continuous spectrum can be written as
(x) =
x

q
2

(1 + x)
rq+(1/4)
(x + )
1/4
_
x(x + )F

+
_
ax + (a + 1)
_
F
_
(35)
where F
3
F
2
(a, b, c; d, e; x) and a, b, c, d, e are given by
a = q +
_
q
2
d = 2 q +
_
q
2
= a + 2
b =
_
q
2
+
_
r
2
g e = 1 + 2
_
q
2

c =
_
q
2

_
r
2
g
(36)
and
= q r 1, g =
1

. (37)
9
In fact it can be shown that expression (35) can be also written in the equivalent
and more explicit form
(x) =
x

q
2

(1 + x)
rq+(1/4)
(x + )
1/4

_
(a +1)
3
F
2
(a, b, c; a +1, e; x) +ax
3
F
2
(a +1, b, c; a +2, e; x)
_
(38)
where no derivatives of
3
F
2
enter. To the best of my knowledge expressions like (38)
with higher-order hypergeometric functions present in the solution of a Schrdinger
equation appear for the rst time in the literature.
To impose the boundary condition at x + (i.e., at z +) we also need
the asymptotic relation
F(x)
x
(d)(e)
(b)(c)
(b a)(c a)
(d a)(e a)
(x)
a
+
(d)(e)
(a)(c)
(a b)(c b)
(d b)(e b)
(x)
b
+
(d)(e)
(a)(b)
(a c)(b c)
(d c)(e c)
(x)
c
(39)
in conjuction with the fact that the rst-order operator M acting on F
Eq. (35) annihilates the term x
a
, for the same reason that M annihilates also
the solution y
0
. Thus, we come to the following conclusions.
A. Scattering states
Denoting by r
R
(k, k

) and r
L
(k, k

) the reection amplitudes from the right and

left respectively, we have
r
R
(k, k

) =
r + i

gk

r i

gk

(2i

gk

)
(2i

gk

( i(k +

gk

))( + 1 i(k +

gk

))
( i(k

gk

))( + 1 i(k

gk

))

2i

gk

(40)
r
L
(k, k

) =
q + ik
q + ik

(1 + 2ik)
(1 2ik)

( i(k +

gk

))( + 1 i(k +

gk

))
( + i(k

gk

))( + 1 + i(k

gk

))

2ik
(41)
where k and k

are the wavenumbers for z and z + respectively.

In other words, we have k =
_
q
2
, k

=
_
(r
2
/g). For the corresponding
reection probability P
r
which is independent of the direction of incidence we
get
P
r
(k, k

) =
cosh
2
(k

gk

) cos
2

cosh
2
(k +

gk

) cos
2

. (42)
B. Bound states
In this case, the requirement for vanishing (x) at innity leads due to (39)
to the three conditions i) a =n, ii) b =n and iii) e c =n. From these we
obtain (see Appendix A for details) the respective regions, red, green and blue, of
10
the phase diagram in Figure 2. More specically, the condition a = n is actually
restricted to the value n = 0 and has a respective (unique) eigenvalue = 0; while
b = n and e c = n lead to the equations (I) and (II) for the green and blue
region respectively.
Note however that in order for solutions of, say, Eq. (I) to exist, it is not sucient
that > 0, but we must also have > f
min
, where f
min
is the minimum value
of the function f() =
_
q
2
+
_
r
2
g, in the left hand side of (I), which is
monotonically decreasing and will therefore cross the horizontal line at height
only if the above inequality holds. This directly implies that the g-interval within
which bound states can exist is
r
2

2
q
2
< g <
r
2
q
2

2
(43)
where it is assumed that the left or the right side of these inequalities will be
replaced by zero or innity, respectively, when the corresponding factor r
2

2
or q
2

2
vanishes or becomes negative. It also follows from (43) that a suitable
range of values for g always exists, so the existence of bound states in the green or
blue region of the phase diagram is always guaranteed. Of course, the inequality
(43) holds also for n i.e., for the full right hand side of (I) (or of (II))
in which case the number of bound states will depend on the value of g and will
generally be smaller than predicted by the simple positivity of the right hand sides
of (I) or (II). Combining the above reasoning with the fact that it will always be
q r 1 < |q| + |r|, leads us to conclude also that = 0 is not a solution of the
eigenvalue equation (I), although it can be a solution of (II) in the middle blue
region; but ultimately even that is rejected for reasons explained in Appendix A.
Hence the only vanishing eigenvalue is the one located in the red region and it
represents the absolute ground state of the system. In the special case whereby the
potentials asymptotic values U
0
and U

are equal i.e., when q

2
= r
2
/g then
clearly f
min
= f()|
=q
2 = 0 and (I) is satised for all positive g, a fact that also
follows directly from the inequality (43). In that case, the eigenvalues are found
easily in closed form and their formula (in the green region, say) becomes
q r < 0
r
q
q r < 1
q r > 1
Figure 2: Phase diagram for bound states
11
in the full hypergeometric case.
Red region: (q > 0, r < 0). The eigenvalue = 0 exists and represents the absolute
ground state of the system.
Green region: (q r > 1). Bound states exist and are determined by the condition

q
2
+

r
2
g = n = q r N (N = n + 1) (I)
Blue region: (q r < 0 for qr > 0 or q r < 1 for qr < 0). Bound states exist and are
determined by the condition

q
2
+

r
2
g = 1 n = r q n (II)
which can be also deduced from (I) by the substitution q q, r r and N n.
Note, however, that in the middle blue region (qr < 0) the value n = 0 is rejected since
then the value = 0, which is now a solution of (II), does not correspond to a physically
acceptable solution (see Appendix A).
White regions: There are no bound states.

n
= q
2

( n)
2
(1 +

g)
2
= q
2

(q r N)
2
(1 +

g)
2
, N = n + 1 = 1, 2, . . . q r. (44)
This will also hold in the blue region upon substituting q q, r r; while the
further substitution N n is also necessary in the outer blue regions.
Let us also mention that the wavefunction of the absolute ground state of the
system we say absolute since there now exist various ground states across the
dierent regions of the phase diagram is given by

0
(x) =
x
q
(1 + x)
rq+(1/4)
(x + )
1/4
. (45)
Other features of the phase diagram worth noting are the following:
a) In contrast to all known examples of solvable potentials, in this case the region
of parametric space where bound states exist is doubly connected. It consists of the
two main regions q r > 1 and q r < 0 with a gap in-between (white regions)
where no bound states exist. b) Every point (q, r), for example in the green-red
region, has a mirror image (q, r) in the middle blue region with the same bound
states except the zero one ( = 0) which exists in the green but not the blue region.
This follows directly from the eigenvalue equations (I) and (II) and the fact that in
the middle blue region the value n = 0 ( = 0) is rejected. Thus we have pairs
of potentials U
+
= U(q, r, x)|
qr>1
and U

= U(q, r, x) with the same bound

states except for = 0 which exists for the former but not the latter potential.
Whether there is some symmetry behind this pairing of potentials in parametric
space is an interesting question that deserves further study.
V. CONFLUENT CASE OF THE FIRST KIND
Let us now study the case where the parameter vanishes, so the solutions (see
Appendix A) are hypergeometric functions of the conuent kind i.e.,
2
F
2
,
1
F
2
etc.
and their behavior at innity will vary accordingly, including now not only powers
of x but also exponentials. We will start with the conuent case of the rst kind,
12
for which = 0 but = 0. The expression of the potential in this case emerges
readily from (33) and (34) upon taking the limit 0, which yields
U(x) =
Ax
4
+Bx
3
+Cx
2
+Dx +E
(x + )
3
(46)
with
A =
1
4

2
B =
1
2

2
+ q
C =
_
q
2
+
1
4

2
+ 2q q +
1
2
+
3
16
_
D =
2
_
2q
2
+ q q +
1
2

1
4
_
E =
3
q
2
(47)
and has a typical plot as in Figure 3.
U(z)
U
0
z
Figure 3: Typical potential of the conuent family of the rst kind. At the potential
tends to a xed value and at + it rises as a harmonic oscillator.
Let us note here that some of the results for the conuent case derive directly from
the previous ones by setting = 0, provided they do not represent a qualitative
change in the problem, in which case it is safer to redo the calculation. Let us
simply mention these results without elaborating much on them. We begin from
the relation z = z(x), for which we have
z

(x) =
_
1 + (x/)
x
x(z)
z
e
z
, x(z)
z+

4
z
2
. (48)
The general expression of the wavefunctions is now
(x) =
x

q
2

e
x/2
(x + )
1/4
_
x(x + )F

+
_
ax + (a + 1)
_
F
_
, (49)
where F(x)
2
F
2
(a, b; c, d; x) and
a = q +
_
q
2
c = 2 q +
_
q
2
(= a + 2)
b = q +
_
q
2
+ p + 1 d = 1 + 2
_
q
2

(50)
13
with q as before formula (30
a
) and
p =
1

(51)
where the new parameter p takes now the position of r in the phase diagram that
we will present shortly. The search for physically acceptable solutions depends
crucially on the sign of , or p, and is based on the asymptotic relation
F(x)
x
(b a)(c)(d)
(b)(c a)(d a)
(x)
a
+
(a b)(c)(d)
(a)(c b)(d b)
(x)
b
+
(c)(d)
(a)(b)
(x)
a+bcd
e
x
. (52)
The results we obtain are as follows.
A. Scattering states
Since the only scattering that makes sense now is from the left, we need only
calculate the reection amplitude r
L
(k) = r(k), which becomes
r(k) =
q + ik
q + ik
(1 + 2ik)(1 q + p ik)
(1 2ik)(1 q + p + ik)
()
2ik
( < 0)
r(k) =
q + ik
q + ik
(1 + 2ik)(q p ik)
(1 2ik)(q p + ik)

2ik
( > 0)
(53)
with |r(k)| = 1 of course, since the form of the potential does not allow for the
particle to pass through towards positive innity.
B. Bound states
A similar calculation as before yields now apart from the vanishing eigenvalue
in the region q > 0, p < 0, which is found separately (see Appendix A) the two
complementary conditions
_
q
2
= p + q N (p < 0, N 1) (54)
and
_
q
2
= p q n (p > 0, n 0) (55)
from which we can readily obtain the results summarized in the next phase diagram.
Clearly, the phase diagram has the type of mirror symmetry we mentioned before
and this raises similar questions as to its origin and interpretation.
Concerning the absolute ground state of the system i.e. the one with = 0
this exists only in the red region and its wavefunction is written as

0
(x) =
x
q
e
x/2
(x + )
1/4
. (56)
Note also that the bound states do not depend on the full set of dimensionless
parameters , , q of the potential U(x) (in the invariant parametrization as before)
but only on their two combinations p(= 1/) and q. This is contrary to what
happens in the scattering data formulas (53) where all three parameters enter.
Something similar to this occurs in the full hypergeometric case we examined before.
The presence of isospectral orbits in parametric space is thus one more feature of
the potentials presented here which has to be explained.
14
p
1
q
p
q + 1
q
2
p
q 1
q
2
p
q 1
q
2
p
1
q
1
Figure 4: Phase diagram for bound states
in the conuent case of the rst kind.
Red region: (q > 0, p < 0). The eigenvalue = 0 exists and represents the absolute
ground state of the system.
Green region: (p < 0 : p (q 1)/q
2
). Bound states exist, and they are given by the
formula
N =

pq
1
2

pN +

pq
1
2

2
+ pN
p
2
, N = 1, 2, . . . , (1 pq)q (I)
Blue region: (p > 0 : p 1/q for pq > 0 and p (q + 1)/q
2
for pq < 0). Bound states
exist, and they are given by the formula
n =

pq
1
2

+ pn +

pq
1
2

2
pn
p
2
, n = 0, 1, . . . (pq 1)q (II)
which can be also derived from (I) by the substitution p p, q q and N n.
But in the left blue region (pq < 0) the value n = 0 is rejected for the same reasons as
before.
White regions: No bound states exist.
VI. CONFLUENT CASE OF THE SECOND KIND
The expression for the potential U(x), in the invariant parametrization, emerges
directly from (46), (47) upon setting = 0, so there will be
U(x) =
_
q
2
q +
3
16
_
x
2
+
_
2q
2
q
1
4
_

2
x + q
2

3
(x + )
3
(57)
with x = x(z) as in (48), which implies an exponentially fast approach to the
asymptotic value U
0
= q
2
of the potential at z , but a very slow decrease to
zero of the kind U 1/z
2
at z +. Depending on the range of values of q
the potential U(z) can take one of the following forms:
15
U(z)
U
0
= 4
z
U
0
=
q

=
1
8
q

=
1
3
q

= 2

1
64
U
0
=
1
9
z z
U(z) U(z)
Figure 5: Representative forms of the potential U(z)
in the conuent case of the second kind.
The wave functions will now be written as (Eq. (49) with = 0)
(x) =
x

q
2

(x + )
1/4
_
x(x + )F

+ (ax + (a + 1))F
_
(58)
where F(x) =
1
F
2
(a; b, c; x/) and
a = q +
_
q
2
, b = a + 2, c = 1 + 2
_
q
2
(59)
while the asymptotic behavior of F() for large || where = x/ in our case
will be given by
F()
||
(b)(c)
(b a)(c a)
()
a
+
(b)(c)
2

(a)
()

_
e
i(+2

)
+e
i(+2

)
_
_
=
1
2
_
a b c +
1
2
__
(60)
e.g., by a linear combination of the three asymptotic behaviors at innity, which
now have the form of two exponentials and one power of due to the fact that the
maximum dimension component of the operator L in (22) with = = 0 is
now of rst order.
On the basis of the above the results we obtain are as follows.
For > U
0
there are scattering solutions both from the left and from the right,
and the corresponding reection amplitudes are given by the formulas
r
L
(k) =
q + ik
q + ik
(1 + 2ik)
(1 2ik)
_

_
2ik
e
2k
, r
R
(k) = ie
2k
(61)
where it is noteworthy that, apart from a constant phase (independent of k), the
reection amplitude from the right is real.
As for bound states, there is only the eigenvalue = 0 with a corresponding
eigenfunction

0
(x) =
x
q
(x + )
1/4
(62)
16
which will satisfy the boundary conditions (0) = () = 0 only if 0 < q < 1/4.
Nevertheless, (62) is not square integrable as one might have expected, due to the
position of the corresponding eigenvalue at the threshold of the continous spectrum.
It hardly needs mentioning, that all the above results irrespective of how
they were produced can be veried directly with a straight substitution in the
Schrdinger equation. To do that, it is advisable to cast that equation in the equiv-
alent x-form
1
w(x)

(x)
w

(x)
2w
2
(x)

(x) +
_
U(x)
_
(x) = 0 (63)
where the derivatives and the expressions of the functions w and U are with respect
to the initial variable x.
Finally, we note that in both main classes of potentials (full hypergeometric case
and conuent of the rst kind) their graphs do not always assume the typical forms
of Figures 1 and 3 but they show interesting variations similar to those in Figure
5. The question of which of these forms may have some special physical interest
will not be discussed here.
VII. PENDING ISSUES AND GENERALIZATIONS
Let us note rst that our preceding discussion on potentials that are solvable via
hypergeometric functions of third order is not exhaustive. The rst reason for this
is that the solution y
0
= x
s
(x+), which we have chosen to reduce the order of the
initial equation, is just a special case for s

= 1 of the general expression, y

0
=
x
s
(x + )
s

, for the solutions of the rst order equation My = 0. This special case
is undoubtedly the most important, for reasons that may have become obvious by
now. But there are also other possibilities e.g., s

= 2 belonging to the terminating

series category and which can thus be solutions of Ly = 0, as required. Given
though that the coecients of the relevant polynomial are now related to one
another (they are all functions of ), the number of independent parameters of the
operator L, and hence of the potential V , will necessarily be reduced, or such a
solution might not even exist.
But there is another reason why our previous discussion is incomplete. We as-
sumed that and are positive and hence, they fall outside the region 0 < x <
between the singular points of the original equation which means that the poten-
tial V is also nite in this region, and therefore also in <z <+ into which
0 < x < is mapped via the transformation z = z(x). But if we assume, for
example, that < 0 and substitute for then x must lie within the inter-
val 0 < x <
1
, which is mapped via the transformation (23) to the seminnite
z-interval < z < 0; and so the corresponding potential V (z) will also be a
half-interval potential with repulsive core of the type 1/z
2
, for z 0. In the same
fashion one needs to examine all the remaining choices of signs for and together
with the corresponding choices of intervals between singular points with respect to
x or z. What is certain, is that even in the context of the third-order case, the
set of solvable potentials is much wider than what we presented here and deserves
further study, especially if such a study could provide answers to some interesting
special questions.
A third pending issue pertains to the comparison of the present family of solvable
potentials with that of Natanzon. What are the dierences or similarities between
the two families? Can one family be deduced from the other by applying the special
17
Darboux transformation related to the removal of the ground state? We will leave
this discussion for a follow-up publication, once we have constructed a much more
general framework of studying solvable potentials and developed a general formal-
ism that is immediately applicable to Natanzon potentials as the simplest special
case. What is certain is that apart from the special case =
1
(for which the
potentials are elementary functions) there is no overlap between the two families,
as evidenced both from the dierent scattering data and the dierent functional
forms of their solutions. (Hypergeometric functions of the type
3
F
2
,
2
F
2
,
1
F
2
in our
case, vs.
2
F
1
,
1
F
1
in the case of Natanzon.)
Next comes the question whether our methodology can be extended to equations
of order higher than third, and produce the corresponding families of solvable
potentials. While this will be the topic of an upcoming publication, we nd it
warranted at this point to present, without proof, some of the pertinent conclusions
in order to underscore the existence of a general framework for the systematic
investigation and enlisting of, potentially, all solvable potentials. Our most basic
results are contained in the following three propositions:
Proposition 1: Every formal eigenvalue equation of the general form
(L + M)y = 0 (64)
where L is any linear dierential operator of order n and M a similar operator of or-
der
n 2, can be reduced to an eigenvalue equation of second order and hence, to
a Schrdinger equation provided that each solution of My = 0 is also a solution
of Ly = 0.
Proposition 2: The weight function w(x) of the Liouville equation that results
from the reduction of any given bidimensional equation (L+M)y = 0 subject to
the constraints of Proposition 1 can always be written in the form
w(x) =
1 + (x/)
x
2
(1 + x)
. (65)
Therefore, the transformation function z(x) that renders the Liouville equation into
Schrdinger form will also be a universal function, i.e., the same function for all
solvable potentials.
Proposition 3: The solvable potentials arising from a bidimensional equation of
an arbitrary order (L+M)y = 0 subject to the constraints of Proposition 1 will
always have the functional form
V (x) =
Ax
4
+ Bx
3
+ Cx
2
+ Dx + E
(x + )
3
(1 + x)
(66)
where the coecients A, B, C etc., are functions of the parameters that remain
free in the operators L and M once we demand that every solution of My = 0 be
also a solution of Ly = 0.
In subsequent publications, we will rely on these three propositions and on some
crucial extensions of them in order to extend this work and develop a systematic
theory of exact solvability of the Schrdinger equation.
VIII. DISCUSSION
18
As was noted at the outset, the critical new idea of this work is this: Given
that the Natanzon search for exactly solvable potentials is exhaustive, we can only
produce new ones if we enlarge the set of functions within which their solutions are
sought for. And if this extended set were the hypergeometric functions of all orders,
then utilizing this set for solving Schrdinger equation would be in principle possible
only in conjunction to the idea of order-reduction. We are thus led naturally to the
notion of eigenvalue equations of higher order of the hypergeometric type if we
want them to be solvable that can be reduced to the Schrdinger equation using
the technique of order-reduction. And as we saw earlier, this idea is indeed feasible
and has actually delivered as a tangible product a new set of solvable potentials,
beyond that of Natanzon. We also saw that this notion provided that propositions
1 through 3 are valid is extensible to a very elegant general formulation which
encompasses, as special cases, both the one studied here where L was of third and
M of rst order as well as the case of Natanzon, with L being of second and M
of zeroth order, i.e., a function of x.
It follows that the method is in principle extensible to equations of higher than
third order, even though it remains to be seen whether the relevant calculations
continue to be manageable. Many further questions arise; here is but a short list of
these:
a) Could there exist an innite hierarchy an innite tower of solvable po-
tentials, with those of the Natanzon class occupying its ground level, the ones we
presented here the rst oor and so on, for the potentials originating from equa-
tions of ever higher order? Or does the process terminate leading to an ultimate
closed set of solvable potentials?
b) What is the origin of the peculiar topology of the phase diagram for bound
states and its mirror symmetry? Are the potential pairs U
+
and U

, connected by
this symmetry, supersymmetric partners? And what about the property of shape
invariance that was shown not to match the initial expectations pertaining to its
range of applicability? In view of the preceding discussion, could it be that an old
conjecture
3
, namely that this invariance may require a wider class of potentials
than that of Natanzon, ought to be investigated further?
c) What is the origin of the reparametrization that reduces the number of active
parameters of the system, or of the isospectral orbits in parametric space? Do
these properties hint at some sort of symmetry that allows also a purely algebraic
approach to the problem?
At a more mathematical level, the mapping of eigenvalue equations of higher
order into the Schrdinger eigenvalue equation a mapping that can be cast in an
elegant compact form could potentially provide valuable new insights to an old
equation which clearly continues to hold an element of surprise.
APPENDIX A
CALCULATION OF EIGENVALUES AND EIGENFUNCTIONS
According to the equations (1) through (10) of section II, we have
(x) = (x + )
7/4
x
q
(1 + x)
rq+(1/4)
_
y
x
s
(x + )
_

(A1)
19
where q and r are dened as in (30) and y(x) is the solution of the bidimensional
equation (22). The latter can be written in terms of the hypergeometric function
3
F
2
as
y(x) = x

3
F
2
(a, b, c; d, e; x) (A2)
where the factor , in the argument = x, serves to transfer the nite singular
point x
0
=
1
of (22) to the standard position
0
= 1 of the hypergeometric
equation, while the factor x

represents the physically acceptable power behavior of

(22) for
x 0. The possible values of (physically acceptable or not) are found as so-
lutions of the unidimensional equation (or Euler-type equation)
L
1
x

= 0 (A3)
where L
1
is the lowest-dimension component of the bidimensional operator L =
L + M of (22). For the power behavior x

at innity the respective equation is

L
2
x

= 0 (A4)
where L
2
is the largest-dimension component of the operator L. (A3) and (A4) are
of course cubic equations for and , which are however easily solved since we
already know (from the known solution y
0
= x
s
(x + )) that one behavior at the
origin is x
s
and one behavior at innity is x
s+1
. The exponents of the remaining
two behaviors are easily found to be
, =
3 s
2

_
V
0
, , =
2 (/) s
2

1

_
V

(A5)
where the specic choice of signs for and (positive for and negative for )
reects the fact that the physically desirable behaviors are then x

and x

for
small and large x respectively. Now we can readily calculate the hypergeometric
parameters a, b, c, d and e in (A2), by recalling that the parameters a
i
and b
j
in
the general hypergeometric series
p
F
q
(a
1
, a
2
, . . . , a
p
; b
1
, b
2
, . . . , b
q
; z) =

(a
1
)
n
. . . (a
p
)
n
(b
1
)
n
. . . (b
q
)
n
z
n
n!
(A6)
have always the same asymptotic meaning: x
ai
(i = 1, . . . , p) are the power be-
haviors of the general solution of the corresponding hypergeometric equation at
innity and x
0
, x
1bj
(j = 1, . . . , q) are the power behaviors at the origin. The
calculation of a, b, c, d and e in (A2) reduces then to simple asymptotic comparison
of the two sides for x 0 and x . The result is
a = s 1, b = , c =
d = s + 1, e = + 1
(A7)
from which, upon using (A5) and the denition of the new parameters q and r (eq.
(30)), we obtain the expressions (36) and the general formula (35) for the solution
(x) that satises the boundary condition (0) = 0 in the region of bound states.
The boundary condition at innity (() = 0) is imposed via the asymptotic
form of (x) for large x which is written as
(x) (a b)Bx

r
2
g
+ (a c)Cx

r
2
g
(A8)
20
and is derived from the asymptotic relation (39) for F(x)
F(x)
3
F
2
Ax
a
+ Bx
b
+ Cx
c
(A9)
by taking into account what we noted earlier: Namely that the rst-order operator
Macting on F (Eq. (35)) annihilates the term x
a
. It follows then from (A8) that
the boundary condition at innity will only be satised if a c = 0 or C = 0. But
the case a c = 0 is rejected at once due to the factor (a c) in the expression
for C, so we are left with C = 0 which can be realized in each of the following ways
i) a = n ii) b = n iii) e c = n (A10)
since the case d c = n is also rejected due to the relation d = a + 2 and the
presense of the factor (ac) in the numerator of C. From (d)/(a) = a(a+1) it
is also clear that for case (i) only the values a = 0 and a = 1 survive, the former
reproducing the already known ground-state solution = 0 (i.e., =
0
) for q > 0,
r < 0, while the latter gives 0 as expected. In case (ii) the pertinent condition
is written as
_
q
2
+
_
r
2
g = q r N (N = 1, 2, . . .) (A11)
and will be satised under the conditions laid out in section IV (Eq. (43)) from
where the exact number of bound states for the problem is deduced. Note that
= 0 is no solution of (A11) for n = 0, i.e., for N = 1, and thus the condition
(ii) yields only positive eigenvalues. In both the above cases, the hypergeometric
function F(x) has polynomial form because both conditions a = n and b = n
are termination conditions for the respective hypergeometric series.
However, in case (iii) contrary to what usually happens for bound states the
hypergeometric series does not terminate since in that case both behaviors x
a
and
x
b
survive at innity, even though only the latter contributes to the asymptotic
limit of (x). The condition for calculating the eigenvalues is now written as
_
q
2
+
_
r
2
g = r q n (n = 0, 1, 2, . . . , r q) (A12)
but especially in the middle blue region of the relevant phase diagram (Fig. 2)
the value n = 0 has to be exempted since then (A12) can have as its solution the
vanishing eigenvalue = 0, which, however, does not lead to a physically acceptable
solution (x) and must be thus rejected. The reason for this is nontrivial. For = 0
a aw arises in the mechanism that ejects the undesirable asymptotic behavior
for x , since then the hypergeometric parameters a, b, etc., no longer depend
on (see Eq. (36)), so the corresponding solution F(x) has no way of knowing
the value of although the relevant ejection operator M= x(x+)+ax+(a+1)
in (35) continues to depend on it.
Using the same techniques as above but with the asymptotic formula (52) in-
stead of (39) we can also calculate the bound states in the conuent case, obtaining
as a result except for the zero eigenvalue the two conditions
_
q
2
= p + q N,
_
q
2
= p q n (A13)
which lead to formulas (I) and (II) of Figure 4.
21
As for the scattering solutions (in the full hypergeometric case for example), we
need the asymptotic form of (z) for z which turns out to be
(z) =
_
z

3/4
(a + 1)e
ikz
z
Be
ik

z
+ Ce
ik

z
(A14)
_
B =
(3/4)
(a b)B, C =
(3/4)
(a c)C
_
and it is clear from (A14) that the scattering solution from the right equals

that is,
R
=

and therefore
r
R
=
_
B
C
_

(A15)
from where after a few steps of algebra we readily obtain the results (40) and
(41). Finally, the calculation of scattering from the left is done by forming the
combination c
1
+ c
2

containing only a traveling wave exp(ik

z) at +, and
proceeding as before. The spectral data for the conuent cases of rst and second
kind are calculated in an analogous manner.
Some closing remarks about the connection of hypergeometric functions and
bidimensional equations are in order. The connection originates from the fact that
bidimensional equations lead to two-term recursion relations, from which the gen-
eral coecient of the series-solution is computed in closed form and it turns out
to be exactly the same as the general coecient of the hypergeometric series (A6).
Thus the hypergeometric functions emerge naturally as solutions of the bidimen-
sional equations of step unity. The equation they satisfy take, for instance, the
function
3
F
2
will naturally have the form (15), but with = 1 (so that the
nonzero singular point lies in the standard position x
0
= 1) and = 0, so that
one exponent of the power behavior at the origin can take the typical value zero
as in the general hypergeometric series (A6) that solves the relevant equation. As
for the exact form of the remaining numerical coecients of (15), as functions of
the hypergeometric parameters a, b, c, d and e, this is never needed since to obtain
the solutions it suces to know the power behaviors of y(x) at zero and innity,
together with the standard asymptotic meaning of the hypergeometric parameters.
As for the hypergeometric functions of the conuent type i.e, with p < q + 1
these emerge naturally from those bidimensional equations whose highest-dimension
component L
2
has no longer the full order of the equation, but is of lower order by
one, two, three, etc., exactly as the value of p compared to q +1. In this case, only
p behaviors at innity where p now stands also for the order of L
2
will be powers
of x; the rest will be exponentials. This alters the character of the point at innity
that now becomes irregular singular. As for the terminology, we have chosen the
terms conuent equation (or function) of the rst kind if p = q, conuent of the
second kind if p = q 1, etc. To complete the picture, we note that not only the
bidimensional equations of step one can be solved via hypergeometric functions, but
also the bidimensional equations of any step . The reason for this is simple. It has
to do with the (rather obvious) fact that the transformations t = x
m
(change of
independent variable) and y = x

Y (change of dependent variable) preserve the

bidimensional character of an equation and merely cause a change in its step, from
to /m, and a shift in its starting powers by , respectively.
22
It thus follows that we will always be able to write the solutions of any bidimen-
sional equation in the form
y(x) = x

q
(a
1
, . . . , a
p
; b
1
, . . . , b
q
; kx

) (A16)
where
p

q
is the general solution of the hypergeometric equation with parameters
a
i
and b
j
, is the step of the given equation and k the numerical coecient that
maps all singular points outside the origin (all of which lie on a circle in the complex
plane) to the standard position = 1 where = kx

. As for the parameters and

a
i
, b
j
in (A16) these are obtained readily from the asymptotic comparison of its
sides at zero and at innity, leading directly to the relations

i
= a
i
,
j
= + (1 b
j
) (A17)
where
j
and
i
are the power behaviors of the given equation at zero and innity
respectively, while of course must be equal to one of the
j
(e.g. =
1
).
It is clear from the above discussion that the hypergeometric functions and
bidimensional equations emerge as the natural framework for the systematic study
of the problem of exact solvability of the Schrdinger equation or any other equation
for that matter. Our future work on this problem will lie within this systematic
framework.
ACKNOWLEDGMENTS
I am grateful to Nick Papanicolaou for helpful discussions and critical comments
and Manolis Antonoyiannakis for helpful discussions and encouragement.
BIBLIOGRAPHY
1
G.A. Natanzon, Vestnik Leningrand Univ. 10, 22 (1971); Teor. Mat. Fiz. 38, 146
(1979)
2
G. Darboux, C.R. Acad, Sci. (Paris) 94, 1456 (1882)
3
F. Cooper, N. Ginocchio and A. Khare, Phys. Rev. D 36, 2458 (1987)
4
F. Cooper, A. Khare, U. Sukhatme, Physics Reports 251, 267 (1995)