J. Phys. A: Math. Gen. 14 (1981) 2645-2651.

Printed in Great Britain

The perturbation of some exactly soluble problems in wave

mechanics by the method of potential envelopes
Richard L Hall and M Satpathyt
Department of Mathematics, Concordia University, Montreal, Quebec, Canada

Received 19 December 1980, in final form 13 April 1981

Abstract. Suppose the function e k ( U ) which represents the kth bound-state eigenvalue of
the Hamiltonian h ={-A+ z)q5(r) + V ( r ) }is known exactly for all allowed values of U P 0.
This article concerns the corresponding eigenvalue &(U) of the Hamiltonian H =
{-A+ uf(q5(r)) + V(r)}, where f(q5) is a smooth, increasing, and either convex or concave
transformation of the potential 4 ( 1 ) .An application of the method of potential envelopes
yields a simple formula for an upper or lower bound to Ek(U)according to whether the
transformation f(Q) is concave or convex. The example q5(r) = (-I-' + W I ) , V ( r )= w 2 r 2 ,
and f(q5) = A-'{eAd - 1) for A > O is discussed in detail.

1. Introduction

Interesting examples of exactly soluble problems in non-relativistic wave mechanics

may be obtained by first choosing the wavefunction and then finding the corresponding
potential (Wigner 1929, Reed and Simon 1978, p 223, Killingbeck 1977b, p 985,
1980). Suppose, for example, we consider wavefunctions with the form $ ( r ) =
Y;"(8,4)r1 exp{-g(r)}, 1 = 0, 1 , 2 , . . , ; then Schrodinger's equation
(-A+ V ( r ) ) $ ( r =
) M r ) (1)
becomes
(E - V(r))r= rg"(r) + 2(1+ l)g'(r) -r(g'(r))2. (2)
If we now choose g to have the form g(r) = i(ur + w r 2 ) , then we can obtain an exact
solution of equation (2) by setting
V ( r ) =v{-(l+1)/r+wr}+w2r2 (3)
and
E = (3 +21)w - v2/4. (4)
The special cases 1 = 0 and w = 0 or z! = 0 correspond respectively to the familiar
hydrogen-like and harmonic-oscillator ground-state problems. Unfortunately it is not
possible to extend this collection of soluble potentials (say by scaling arguments) to
include the potential V ( r )= (-A/? +Br + Cr2)for arbitrary A, B and C 3 0 because the
function g which generates the problem has only two parameters: if the potential in
equation (3) is simply multiplied by a constant, the corresponding exact ground-state
t Permanent address: Physics Department, Utkal University, Bhubaneswar 75 1007, India.

0305-4470/81/102645 + 07$01.50 @ 1981 The Institute of Physics 2645

2646 R L Hall and M Satpathy

energy remains unknown. We have recently developed a general approximation theory

for sums of soluble potentials (Hall 1981) and this can be used to treat those ( A ,B, C )
combinations which have no known exact solutions.
In the present article we use the method ofpotential envelopes (Hall 1980) to show
how one can make use of exactly soluble problems such as those arising from equation
(2) to approximate the eigenvalues of problems which are 'nearby' in a sense rather
different from conventional perturbation theory. In particular we examine potentials of
the form
V ( r ) = uf(-r-' + o r ) +0 2 r 2 (5)
where f is a smooth increasing function of 4 = (-r-' f o r ) . The principal idea of our
geometrical theory is that a tangent line to the curve (q5,f(4))has an equation like
y = a + &b with /3 > 0 and therefore generates the 'tangential potential' defined by
V ( ' ) ( r )= u a + v ~ ( - r - ~ + o r ) + o ~ r ~ . (6)
Consequently, by equations (3) and (4), we know that the Hamiltonian H") =
(-A + V ( ' ) ( r ) )has the exact lowest eigenvalue given by
E") = ucx + 3~ -(~p)'/4. (7)
The method of potential envelopes which is outlined in 92 prescribes sufficient
conditions on f which allow us to relate the unknown eigenvalue E to the tangential
eigenvalues E(').In many cases the method yields a simple algebraic formula, in terms
of the functions f and f', for bounds on E. In 9 3 we return to the potential V ( r ) of
equation ( 5 ) and study the illustration f(4)= (e*' - l ) / h in some detail.

2. The method of potential envelopes

We suppose that the exact eigenvalue & ( U ) corresponding to the lowest bound state of
the Hamiltonian
h^ = -A + uq5 ( r ) + U ( r ) (8)
is known for each allowed value of the positive coupling constant U . The graph ( U , E ( U ) )
we call the energy trajectory of 6 with respect to 4. The new Hamiltonian which we wish
to analyse in terms of h^ is given by
H =-A+vf(4(r))+ U ( r ) (9)
and our aim first of all is to approximate the lowest eigenvalue E of H. We assume that
the transformation function f is continuously differentiable and that f'(q5) > 0, for all
4 ( r ) with r > 0.
The key additional assumption which leads to energy bounds is that f is either
convex or concave: we shall take the case that f is convex for all 4 ( r ) , r > 0 , which leads
to lower energy bounds; upper energy bounds are obtained when f is concave, by an
exactly similar argument.
Consider the point r = s and 4(s)= t which, for the present, we keep constant. We
have, by the convexity off,
f ( # J ( r ) w) ( r ) ( r =
) f ( t )+ f ' ( t ) M ( r )- t ) all r > 0. (10)
Equation (10) simply states that f lies qbove the tangent line f")which touches f at 4 = t.
 The method of potential envelopes 2647

We call the collection of functions { f " ' ( r ) } an envelope representation for the function
g ( r ) = f(4( r ) ) and write this g = E:F$e{f(f)}, where R* is the range of the potential
function 4.
We can rewrite equation (10) (with t still fixed) in the form

f(4( r ) )+)w = a 0 )+ P (t)4( r ) (11)

where the 'constants' a ( t ) and P ( t ) are given by

0)= f ( t )- t f ' ( t ) P ( t ) = f ' ( t )> O (by hypothesis). (12)

We now define the tangential Hamiltonian H") by
H") = -A + ~ f ( ' ) (+r )V ( r ) (13)
and observe that, since f") = a +P4, where a and P are constants, the exact lowest
eigenvalue E")of H") is given in terms of the known energy-trajectory function E ( U ) of
tG by

E") = a v + E ( P v ) . (14)
Since f f ) ( r )< f ( ~ $ ( r ) ) , r > 0, we have, by Weyl's comparison theorem (e.g. Weinstein
and Stenger 1972, p 21), that
E")< E for each fixed choice of t E RQ (15)
where R4 is the range of the function 4. The lower bound ELis now given by
EL = Envelope{E(t))
rsR+ s E. (16)
Our smoothness assumption concerning f has been made merely for the con-
venience of using elementary calculus. In the same spirit, we now assume that f'(t) and
&'(U) are continuous functions so that the envelope in equation (16) can easily be
calculated: our assumption thatf is convex now becomesf"(t) > 0, all t E R+. If we write
equation (14) in the form
G ( v ,E, t ) = a u + E ( P u ) - E= O (17)
and think of this equation as representing a family of curves in the ( U , E ) plane labelled
by the parameter t, then we obtain (e.g. Courant 1936, p 172) the envelope curve ( U , E L )
of this family by eliminating the parameter t between the two equations
aG
G(u, E, t ) = 0 and -(U, E, t ) = O . (18)
at

In our problem these envelope equations become explicitly:

E L= v { f ( t )- tf'(t)}+ E ( u f ' ( t ) )
t = E'(Uf(t)).
If equations (19) were difficult to solve, one might prefer to approach the original
problem directly by using numerical methods. However, in many cases these equations
can be rearranged to give E L and U each as explicit functions of t, that is to say,
parametric equations for the graph ( U , EL).If, for example, + ( r ) is the pure power law
q5(r) = sgn(p)rP,where p is a constant greater than or equal to -1, and V(r)=O,then
&(U) = sgn(p)F'P'v2''pC2', where F ( p )is a positive constant (Hall 1980), and equations
2648 R L Hall and MSatpathy

( 1 9 ) become in this case (after some’algebraic manipulation):

EL=;u{2f(t)+ptf’(t)} pt>O
v = j2F‘P’/t(p+ 2)j(p+2”p{f(t)}-1.
Equations ( 2 0 ) agree, of course, with equation ( 4 . 1 2 ) of Hall ( 1 9 8 0 ) : the actual
Hamiltonian for which ( U , EL)provides a lower trajectory bound is given in the notation
of the present article by
H = -A + uf(sgn(p)rP) (21)
and the equivalence of results is easily verified with the aid of the equations t = sgn(p)sP
and ptf’(t)= s df(sgn(p)sP)/ds.
Of course, equations ( 1 9 ) and ( 2 0 )immediately yield upper trajectory bounds when
f is concaue. Also, whenever the trajectory function & k ( V ) is known for the kth excited
state of the base problem l,the corresponding energy bound formula yields a bound on
the kth eigenvalue Ek of the transformed problem H : this follows because the
comparison theorem of Weyl applies to each eigenvalue separately.

3. An example

In order to illustrate the theory of 0 2 we need first to choose 4 ( r ) and U ( r )so that the
Hamiltonian 6 corresponds to an exactly soluble problem; then we must choose the
convex or concave transformation f(4)of 4 which generates the new Hamiltonian H.
The present paper is concerned with the class of Hamiltonians discussed in § 1 above
where U ( r )= w2r2, We therefore define 6 as follows
6 = -A+ v 4 ( r )+ w 2 r 2
4 ( r )= (-r-’ + wr) (22)
&(U)=3w-U2/4
where ( U , E ( U ) ) is the exact energy trajectory of fi with respect to 4. The Hamiltonian H
which we study then becomes
H = -A+ v f ( - r - ‘ + w r ) + w 2 r Z (23)
where f is increasing and either convex or concave, and will be chosen explicitly later.
By applying the theory of § 2 we find the following approximation E A to the exact
energy trajectory E ( u )of H (equations ( 2 2 ) have been substituted directly in equations
( 1 9 ) and the resulting equations have been rearranged to give U and E in terms of the
parameter t ) :
U = -2t(f’(t)}-l E A= u f ( t )+ t 2 + 3w. (24)
The parameter t must be in the range of 4 which, in this example with w # 0, is the entire
real line; moreover, since f is assumed to be increasing and U is positive, we see that
t S 0 . This means that, in this example, we only make use of that subset of the
enveloping family of curves {a+ & 5 ( r ) }corresponding to points of contact r = s s
U - ’ ” : the rest of the enveloping family gives suboptimal energy estimates.
We now choose the transformation f of 4. For the identity f(4)= 4, we see from
equation ( 2 4 ) that E A= E ( U ) , as we expect. If we choose for f a function which is
 The method of potential envelopes 2649

convex, then equation (24) will yield a lower trajectory bound which we can compare
with an independent variational upper bound. The example we present is
f(4)= A-'{eA' - 1) A>O (25)
which approaches the identity f ( 4 )= 4 as A & 0. The actual potential generated in this
way (with A > 0) has the form
V(r) = vA -'{exp[A(-r-' + wr)] - 1) + w2r2. (26)
A sketch of V(r) for the case v = 2 and w = 20A = 1 is shown in figure 1: for A > 0, the
Coulomb singularity of 4 ( r ) is removed by the transformation f ( 4 ) for we have
-
V(r) 5 lim,ro V(r) = -vA-'; for large values of r, V(r) vh-' exp(Awr). The (lower)
trajectory bound equation (24) becomes (in parametric form)
v = 2q eAq q30
EL=uA-'{e-A4-l}+q2+3w
where we have used the positive parameter q = -t. Thus we have in equation (27) a
formula for a lower bound to the lowest eigenvalue of the Hamiltonian H = {-A + V(r)}
which depends on v 5 0, w 3 0 and A > 0; in the limiting case A & 0, this formula yields of
course the exact result E = 3w - v2/4. For w = 1, the graph of EL(v)for A = 0.05 is
shown in figure 2 together with a variational upper trajectory bound E U ( v )found by

Figure 1. The potential in the example is the Figure 2. Upper and lower bounds to the lowest
smooth transformation uf(4) + w z r z of the eigenvalues E ( u ) of the Hamiltonian H =
potential 4 ( r )= (-I-' + w r ) , in which f(4)= - A + v A - ' [ e x p { A ( - r + ' + w r ) } - l ] + w Z r 2 , with w = 1
A-'{exp(A4)- 1). Explicitly we have V ( r )= and A =0.05: in the limit as ALO, we have H =
wA-'[exp{A(-r-'+wr)}- 1 ] + w 2 r 2 , and the figure -A+ u(-r-'+wr)+w2r2 and E ( u )= 3w - v2/4
illustrates the case w = 20A = 1, and U = 2. exactly.
2650 R L Hall and M Satpathy

using the trial function $ ( r ) = exp{-ar - br’} and minimising the expectation of H with
respect to the scale variables a and b.
In order to find EL for a given v numerically, one first solves equation (27a) for q
and then substitutes this value of q into (27b). For the case w = 1, U = 5, and
A = 0.01, 0.05 and 0.1 Killingbeck (1981) has found a numerical approximation
EK to E by the method described in his (1977a) article: these results
together with ours are respectively (EL, E K ,E u ) = (-3.099, -2.917, -2.908),
(-2.578, -1.971, -1.950), and (-2.066, -1.230, -1.160).
Of course, our lower bound equation (27) gives the general form of the dependence
of E on the triple ( v , A, w ) and this analytical information, which is always interesting, is
particularly important for the N-body problem, N 2 2 , as we have shown in Hall
(1980, 1981). Even for the rather general class of potentials (23), we have a simple
recipe (24) which yields lower bounds when f” > 0 and upper bounds when f” < 0.

4. Conclusion

The minimal properties of the eigenvalues of the semi-bounded self -adjoint operators
of wave mechanics allow us to use the geometrical ideas of convexity and envelope
representations in their study. In this article we submit a component 4 ( r ) of the
potential in the Hamiltonian 6 to a smooth convex or concave transformationf(4). This
generates a new Hamiltonian H whose kth eigenvalue can be easily approximated
provided we know how the kth eigenvalue of 6 depends on the coupling constant v of
the component 4 ( r ) , that is to say, the kth energy trajectory of 6 with respect to 4. In
terms of perturbation theory one can also write

We do not, however, require v to be small: provided f is convex or concave the method

always yields a strict energy bound ; whenever the curvature of f is small in the region
over which the square of wavefunction l$(r)I’ is significantly large, the approximation
will be good. The purpose of this geometrical theory is to provide a new tool with which
to analyse potentials that are near to one of the few potentials we have whose
corresponding energy trajectories are known exactly.

Acknowledgments

We should like to thank J Killingbeck and B Simon for their comments on an earlier
draft of this paper, and the Natural Scien’ces and Engineering Research Council Canada
for partial support of this work by Grant number A3438.

References

Courant R 1936 Differential and Integral Calculus vol 2 (London: Blackie)

Hall R L 1980 Phys. Rev. D 22 2062-72
-- 1981 Phys. Rev. D 23 1421-9
 The method of potential envelopes 265 1

Killingbeck J 1977a J. Phys. A : Math. Gen. 10 L99-103

-1977b Rep. Prog. Phys. 40 963-1031
-1980 J. Phys. A: Math. Gen. 13 L393-5
-1981 private communication
Reed M and Simon B 1978 Methods of Modern Mathematical Physics I V : Analysis of Operators (New York:
Academic)
Weinstein A and Stenger W 1972 Methods of Intermediate Problems for Eigenvalues (New York: Academic)
Wigner E P 1929 Z. Phys. 30 465-7