Heun’s equation with nearby singularities

By W o l f g a n g L a y a n d Sergei Y u. Slavyanov
Universität Stuttgart, Institut für Theoretische und Angewandte Physik,
Pfaffenwaldring 57, 70550 Stuttgart, Germany

Received 19 October 1998; revised 8 March 1999; accepted 28 May 1999

Dedication. Three years ago Professor Felix M. Arscott died. We miss him as a
charming gentleman, a good friend and a brilliant teacher in mathematics. He pre-
sented his wide knowledge on special functions with utmost clearness and he was
very patient with those whom he communicated. We took part in writing a book on
Heun equations together with him. He was the author of the basic chapter on Heun
equations (Heun differential equations, 1995, OUP). This article may be regarded as
an addendum to this chapter and we dedicate it to F. M. Arscott.
The boundary problem for the Heun equation is studied in the case when two singu-
larities of the equation merge. The eigenfunctions are characterized by a sharp peak
in a small boundary layer around the merging points. The eigenvalues as functions
of the distance between the merging points reveal a non-analytic behaviour in the
vicinity of zero.
Keywords: Heun equation; Fuchsian equation; asymptotics;
connection problem; eigenvalue curves

1. Introduction
The problem studied in this paper is as follows. Consider the Heun equation—the
Fuchsian equation with four singularities—in the canonical form
Lz y(z) = λy(z) (1.1)
with
Lz = z(z − 1)(z + s) D2 + [c(z − 1)(z + s) + dz(z + s) + ez(z − 1)] D + abz
and
d
D=
.
dz
The parameters a, b, c, d, e are not independent of one another. The following linear
relation holds (Fuchs identity):
a + b + 1 − c − d − e = 0. (1.2)
It is assumed that a, b, c, d, e are real and without loss of generality satisfy the fol-
lowing conditions:
c
1, d
1, e
1, a
b, (1.3)

Proc. R. Soc. Lond. A (1999) 455, 4347–4361 c 1999 The Royal Society


4347
4348 W. Lay and S. Yu. Slavyanov

resulting in the existence of one analytical solution and one unbounded solution
at each regular singularity z = 0, z = 1, z = −s of (1.1). As a consequence of
equations (1.2), (1.3) we have
a + b
2, a + b − d
1. (1.4)
Suppose that the parameter s is positive and is small, i.e. s  1.†
We are going to study the local behaviour of the solutions of (1.1) near its sin-
gularities. For this reason we use the so-called ‘Riemann scheme’ (RS for short; see
Slavyanov et al . (1997)) which in the case of equation (1.1) reads
 
−s 0 1 ∞ ;z
 0 0 0 a λ. (1.5)
1−e 1−c 1−d b
The locations of the singularities are written in the first row and the characteristic
exponents of the Frobenius solutions at the corresponding singularities are written
in the second and third rows. The last column shows the independent variable z and
the accessory parameter λ. We stress that this parameter does not affect the local
behaviour of the solutions of (1.1) at its singularities.
A boundary problem (central two-point connection problem) on the interval [0, 1]
is posed for equation (1.1) with the boundary conditions
|y(0)| < ∞, |y(1)| < ∞. (1.6)
Eigenvalues λn of this problem are functions of the parameter s. The question we
are dealing with is: what is the behaviour of the eigenvalue curves λn = λn (s) in
the vicinity of s = 0, i.e. for s → 0? The qualitative behaviour of the eigenfunctions
yn (z) is also studied.
At first glance this seems to be a typical problem in the asymptotic theory of
ordinary differential equations since two singularities merge.¶ However, the answer
cannot be found in standard textbooks (see, for example, Fedorjuk 1993; Slavyanov
1996). The reason for this is that the standard procedure valid for two merging
turning points fails in the case of two poles.
The problem of two merging poles was first studied by Abramov & Slavyanov
(1978) in the physical context of the two Coulomb centres problem when two elec-
trical charges come close to one another. The underlying idea of solution in the
problem considered here is the same. However, the calculations and the results are
different. A physical example studied by numerical computations in Lay (1991) is
discussed at the end of the article. The theory presented here is necessary to explain
the calculations carried out in Lay (1991).
The principal steps of the proposed method include the use of different scales
(boundary-layer method), perturbation expansions, matching procedures and the
method of successive approximations in solving the characteristic equation for the
eigenvalues.
† In publications on Heun’s equation, −s usually stands instead of s but our choice will be more
convenient for the following computations.
¶ It is necessary to stress that merging of singularities should be distinguished from confluence. In
the latter case singularities not only coalesce but, moreover, some of the parameters of the considered
equation tend to infinity (see Seeger et al. 1995).

Proc. R. Soc. Lond. A (1999)

 Heun’s equation with nearby singularities 4349

First, on a formal level we study the strictly limiting case. Setting s = 0 in (1.1)
and defining λ0 = λ|s=0 = ρ1 (ν)ρ2 (ν), where the unknown quantities ρ1 (ν), ρ2 (ν)
satisfy the condition (Fuchs relation)
ρ1 (ν) + ρ2 (ν) = d − a − b, (1.7)
we obtain

Lz → L0z , 

L0z = z{z(z − 1) D2 − [a + b + 1 − d − (a + b + 1)z] D + ab}, (1.8)


L0z yn0 = ρ1 (ν)ρ2 (ν)yn0 (z).
Equation (1.8) yields
yn0 = z ρ1 F (a + ρ1 , b + ρ1 , 1 + ρ1 − ρ2 ; z) (1.9)
as one of the solutions, where F (α, β, γ; z) is the standard hypergeometric function
(see, for example, Olver 1997). Unless no specific values of the parameters are chosen
the other solution is obtained by the replacement† ρ1 → ρ2 , ρ2 → ρ1 in (1.9).
The quantities ρ1 (ν), ρ2 (ν) are parametrized by ν and this parametrization refer-
ring to the Fuchs condition (1.7) may be taken as
ρ1 (ν) = ν, ρ2 (ν) = d − a − b − ν. (1.10)
For the sake of brevity this dependence of ρ1 , ρ2 is normally omitted below. Our goal
is to obtain eigenvalues λn and in the above-defined terms to construct an eigenvalue
equation for ν.

2. Forms of solutions
By splitting the terms of the zeroth order of magnitude and of the order of magnitude
proportional to the small parameter s we present the operator Lz in the form
Lz = zMz0 + sMz1 (2.1)
with
Mz0 = z(z − 1) D2 − (a + b + 1 − d − (a + b + 1)z) D + ab,
(2.2)
Mz1 = z(z − 1) D2 + (−c + (d + c)z) D.
Along with equation (1.1) an auxiliary equation is studied, namely
[zMz0 u(z) − ρ1 ρ2 ]u(z) = 0, (2.3)
where the role of the parameter ν is discussed below. Solutions of equation (2.3)
bounded at the singularity z = 1 are denoted by uν (z). The RS exhibits the local
behaviours of the solutions related to equation (2.3) and reads
 
0 1 ∞ ;z
ρ 1 0 a . (2.4)
ρ2 1 − d b
† Here and below no attention is paid to specific values of the parameters where more refined formulae
for the solutions of the hypergeometric equation would be needed.

Proc. R. Soc. Lond. A (1999)

4350 W. Lay and S. Yu. Slavyanov

It is straightforward to calculate that the functions uν (z) are expressible in terms of

the hypergeometric function F (α, β, γ; z) (see Olver 1997) as
uν (z) = z ν F (a + ν, b + ν, d; 1 − z). (2.5)
Further on, it will be computed that, in the first approximation, ν ∼ n where n ∈ Z.
As it follows from (2.4) and (1.10) the value of ν reveals the branching behaviour
of the recessive solution of equation (2.3) at zero as it is one of the characteristic
exponents.
We represent the eigenfunctions yn (x) of the boundary problem (1.1), (1.6) by
means of a formal series
m=∞
yn (z) = gnm uν+m (z) with gn0 = 1. (2.6)
m=−∞

This ansatz with fixed normalization is valid in the interval [, 1] where  = o(1) as
s → 0. We denote it by yn+ .
Another series is needed in a neighbourhood of zero where a new scale for the
independent variable is introduced:
z := ζs. (2.7)
Thus, instead of (1.1) we get
Lζ y(ζ) = λy(ζ) (2.8)
with
Lζ := ζ(ζ + 1)(ζs − 1) D2 + [c(sζ − 1)(ζ + 1) + dsζ(ζ + 1) + eζ(sζ − 1)] D + absζ
and
d
D= .
dζ
The differential operator in equation (2.8) can be split into the sum of two operators:

Lζ = Kζ0 + sζKζ1 , 

0 2
Kζ = −ζ(ζ + 1) D + [−c − (a + b + 1 − d)ζ], (2.9)


1 2
Kζ = ζ(ζ + 1) D + [d + c + (a + b + 1)ζ] D + ab.

In addition to the auxiliary equation (2.3) we introduce a second auxiliary equation

valid for ζ = O(1):
(Kζ0 v(ζ) − ρ1 ρ2 )v(ζ) = 0. (2.10)
The RS of this equation is given by
 
−1 0 ∞ ;ζ
 0 0 −ρ1 . (2.11)
1 − e 1 − c −ρ2

Proc. R. Soc. Lond. A (1999)

 Heun’s equation with nearby singularities 4351

We denote solutions of (2.10) bounded at zero by vν (ζ). They may be expressed in

terms of the hypergeometric function according to
vν (ζ) = F (−ρ1 , −ρ2 , c; −ζ). (2.12)
The eigenfunctions yn (z) of the boundary problem (1.1), (1.6) are represented in the
form of a series:
m=∞
yn (z) = hnm vν+m (ζ). (2.13)
m=−∞

This series as a formal ansatz is assumed to fit the problem at z = o(s−1 ) and we

denote it by yn− . In § 3 we study the perturbation procedure with the help of which
the coefficients gnm and hnm of the series (2.6) and (2.13) are found recursively.
Thereafter, a matching procedure allows us to obtain the unknown characteristic
exponent ν as well as the eigenvalues λn .

3. Perturbation procedure
The asymptotic procedure of getting the succeeding coefficients gnm , hnm in expan-
sions (2.6), (2.13) is based on the formal series
∞
gnm = s|m|+j gnmj , (3.1)
j=0
∞
hnm = s|m|+j hnmj . (3.2)
j=0

First, we find difference equations for the coefficients gnm and hnm and further on
the perturbation procedure for obtaining gnmj and hnmj . We start with the case of
hnm . Consider the operators
Z : Zf (x) = xf (x), T : T f (x) = x(x + 1)Dx f (x) (3.3)
obeying the following commutation relation:
[T, Z] = T Z − ZT = Z(Z + 1). (3.4)
In terms of these operators,
ζKζ1 = −ζKζ0 − dTζ + abZζ (3.5)
holds. The operators Z and T initiate the well-known three-term recurrence relations
between the functions vν+j (z):
j=1
Zvν (ζ) = fν,j vν+j (ζ), (3.6)
j=−1
j=1
T vν (ζ) = eν,j vν+j (ζ), (3.7)
j=−1

Proc. R. Soc. Lond. A (1999)

4352 W. Lay and S. Yu. Slavyanov

with

c(ρ1 + ρ2 − 1) + 2ρ1 ρ2 
fν,0 = − ,

(ρ1 − ρ2 − 1)(ρ1 − ρ2 + 1) 




cρ1 (ρ2 + c)
fν,−1 =− , (3.8)
(ρ1 − ρ2 − 1)(ρ1 − ρ2 ) 



ρ2 (ρ1 + c) 

fν,1 =− , 

(ρ1 − ρ2 + 1)(ρ1 − ρ2 )


cρ1 ρ2 (2c − a − b + d − 1) 
eν,0 = − ,

(ρ1 − ρ2 − 1)(ρ1 − ρ2 + 1)  



cρ1 ρ2 (ρ2 + c)
eν,−1 = , (3.9)
(ρ1 − ρ2 − 1)(ρ1 − ρ2 ) 



ρ1 ρ2 (ρ1 + a) 

eν,1 = . 
(ρ1 − ρ2 + 1)(ρ1 − ρ2 + 1)
Combining (3.5)–(3.7) we obtain
j=1 j=1
ζKζ1 vν (ζ) = −(ρ1 + a)(ρ2 + a) fν,j vν+j (ζ) − d eν,j vν+j (ζ). (3.10)
j=−1 j=−1

It comes to a three-term recurrence system for the coefficients hnm : substituting

series (2.13) into equation (1.1) taking into account (2.7), (2.9), (3.10) yields

s[(−ρ1 − a + m)(ρ2 + a + m)fν+m−1,1 − deν+m−1,1 ]hn,m+1 − [(−ρ1 + m)(ρ2 + m) + λn

− s[(ρ1 − a + m)(ρ2 + a + m)fν+m,0 − deν+m,0 ]hnm
+ s[(ρ1 − a + m)(ρ2 + a + m)fν+m+1,1 − deν+m+1,1 ]hn,m−1 = 0. (3.11)
The eigenvalues λn (ν) are expanded in a formal series
∞
λn (ν) = sj λnj (ν). (3.12)
j=0

On the first step of the recursive procedure, λn0 (ν) is obtained as

λn0 (ν) = ρ1 (ν)ρ2 (ν). (3.13)
On the next step the coefficients hn,±1,0 are obtained, then λn1 (ν) and so on. Each
time the succeeding terms with the same power of s are cancelled.
For the coefficients gnm the procedure is analogous and a recurrence system similar
to equation (3.11) can be constructed.

4. Matching procedure to zeroth order

In order to match the constructed solutions (2.6) and (2.13) we use standard formulae
connecting the local solutions of the hypergeometric equation (see Bieberbach 1965).

Proc. R. Soc. Lond. A (1999)

 Heun’s equation with nearby singularities 4353

For instance it holds that (see Bieberbach 1965, eqn (8.8.4))

F (α, β, γ; z)
Γ (γ)Γ (γ − α − β)
= F (α, β, 1 + α + β − γ; 1 − z)
Γ (γ − α)Γ (γ − β)
Γ (γ)Γ (α + β − γ)
+ (1 − z)γ−α−β F (γ − α, γ − β, 1 + γ − α − β; 1 − z).
Γ (α)Γ (β)
(4.1)
Replacing z → 1 − z, and γ → 1 − γ + α + β in (4.1) gives
F (α, β, 1 − γ + α + β; 1 − z)
Γ (α + β + 1 − γ)Γ (1 − γ)
= F (α, β, γ; z)
Γ (1 + β − γ)Γ (1 + α − γ)
Γ (α + β + 1 − γ)Γ (γ − 1) 1−γ
+ z F (1 + β − γ, 1 + α − γ, 2 − γ; z). (4.2)
Γ (α)Γ (β)
We now take (2.5) and identify it with (4.2); this yields
α = a + ρ1 , β = b + ρ1 , γ = 1 + ρ1 − ρ2 . (4.3)
+
From (2.6) and (4.3) it follows that the asymptotic behaviour of functions y (which
are bounded at z = 1) in the vicinity of z = 0 is
Γ (d)Γ (ρ2 (ν) − ρ1 (ν)) ρ1 (ν)
y+ ∼ z [1 + o(z)]
Γ (b + ρ2 (ν))Γ (a + ρ2 (ν))
Γ (d)Γ (ρ1 (ν) − ρ2 (ν)) ρ2 ν
+ z [1 + o(z)] as z → 0. (4.4)
Γ (b + ρ1 (ν))Γ (a + ρ1 (ν))
It also holds that (see Bieberbach 1965, eqn (8.6.14))
Γ (γ)Γ (β − α)
F (α, β, γ; z) = (−z)−α F (α, 1 + α − γ, 1 + α − β; 1/z)
Γ (β)Γ (γ − α)
Γ (γ)Γ (α − β)
+ (−z)−β F (β, 1 + β − γ, 1 + β − α; 1/z). (4.5)
Γ (α)Γ (γ − β)
We take formula (4.5) and apply it to (2.12), (2.13) getting
Γ (c)Γ (ρ1 (ν) − ρ2 (ν)) ρ1 (ν)
y − ∼ hn0 ζ [1 + O(ζ −1 )]
Γ (−ρ2 (ν))Γ (c + ρ1 (ν))
Γ (c)Γ (ρ2 (ν) − ρ1 (ν)) ρ2 (ν)
+ hn0 ζ [1 + O(ζ −1 )] as ζ → ∞. (4.6)
Γ (−ρ1 (ν))Γ (c + ρ2 (ν))
y − is bounded at z = 0 and formula (4.6) represents it near z = 1.
In the overlapping interval where both asymptotic expansions (4.4) and (4.6)
√ are
valid (see√figure 1) the magnitude of z and of ζ may be taken as z = O( s) and
ζ = O(1/ s).†
† Other possible choices make practical computations in higher orders of powers of s more compli-
cated. In the first-order approximation, evidently, the calculations as well as the results do not depend
upon the intermediate scaling.

Proc. R. Soc. Lond. A (1999)

4354 W. Lay and S. Yu. Slavyanov

z
−s 0 1

Range of common validity of O(s)−1/2

−1 0 1/s ζ

Figure 1. Range of common validity of y + and y − .

Therefore, it is appropriate to introduce the matching variable t according to

√ √
z = t s and ζ = t/ s.
From (4.4), (4.6) we get

y + ∼ Atρ1 (ν) + Btρ2 (ν) ,

(4.7)
y − ∼ Ctρ1 (ν) + Dtρ2 (ν) ,

with the following symmetry relations for A(a, b, d, ρ1 , ρ2 ), B(a, b, d, ρ1 , ρ2 ), C(c, ρ1 ,

ρ2 ), D(c, ρ1 , ρ2 ):
A(a, b, d, ρ1 , ρ2 ) = B(a, b, d, ρ2 , ρ1 ), C(c, ρ1 , ρ2 ) = D(c, ρ2 , ρ1 ).
The matching condition is the zero of the Wronskian of y − and y + , which yields
AD − BC = 0. (4.8)
If actual values of A, B, C, D are substituted, the matching condition (4.8) results in

Γ (d)Γ (c)Γ 2 (ρ2 (ν) − ρ1 (ν)) ρ2 (ν)

s (1 + O(s))
P (a, b, c, ρ2 )Γ (−ρ1 (ν))
Γ (d)Γ (c)Γ 2 (ρ1 (ν) − ρ2 (ν)) ρ1 (ν)
− s (1 + O(s)) = 0, (4.9)
P (a, b, c, ρ1 ) Γ (−ρ2 (ν))
where
P (a, b, c, ρ) = Γ (a + ρ)Γ (b + ρ)Γ (c + ρ).
This leads to the characteristic equation for the values of ν:
Γ 2 (ρ2 (ν) − ρ1 (ν))Γ (−ρ2 (ν)) P (a, b, c, ρ1 (ν))
= sρ1 (ν)−ρ2 (ν) (1 + O(s)). (4.10)
Γ 2 (ρ1 (ν) − ρ2 (ν))Γ (−ρ1 (ν)) P (a, b, c, ρ2 (ν))

Proc. R. Soc. Lond. A (1999)

 Heun’s equation with nearby singularities 4355

Since s is a small parameter, the left-hand side of this equation is also a small
quantity. This means that one of the gamma functions in the denominator of the
expression in the left-hand side must be large. Suppose that
ν = n + δ, (4.11)
where
|δ|  1. (4.12)
Then, according to equation (1.10), the multiplier Γ (−ρ1 ) in the denominator of
the expression in the left-hand side of equation (4.10) is large at small values of δ.
Expanding the functions on both sides of (4.10) about δ = 0, we obtain
(−1)n
δ = sa+b−d+2n × G0 (a, b, c, d, n) (4.13)
n!
with
1 Γ 2 (ρ01 − ρ02 ) P (a, b, c, ρ02 )
G0 (a, b, c, d, n) = , (4.14)
0
Γ (−ρ2 ) Γ 2 (ρ02 − ρ01 ) P (a, b, c, ρ01 )
where
ρ01 = n, ρ02 = d − a − b − n. (4.15)
Tacitly, we have assumed that in (4.13) none of the quantities a, b, c, a + b − d is an
integer. In this case only the choice for ν written in (4.11) is consistent with the other
assumptions in the problem. Otherwise, instead of simple roots of equation (4.10),
double-degenerate roots may occur due to arguments of two gamma-functions being
close to their poles. This case needs a special treatment. The leading term of the
eigenvalue asymptotics in our case, here, according to equations (3.13), (4.15) is
λ0n0 = n (d − a − b − n). (4.16)
The impact to the eigenvalue ∆λn due to the boundary layer in the neighbourhood
of z = 0 may be calculated with the help of the formula
∂λ
∆λn = δ + O(δ 2 ). (4.17)
∂ν
In the first approximation we obtain
(−1)n
λn = λn0 (1 + O(s)) − s2n+a+b−d (2n + a + b − d) × G0 (a, b, c, d, n). (4.18)
n!
The second term according to condition (1.4) is smaller compared with the order
of magnitude of the first term. However, according to the philosophy of exponential
asymptotics† it should be retained. It is rapidly decreasing when the number n
increases and is small from the beginning if a + b − d is large.
As mentioned above, the boundary-layer phenomenon causes a large peak of the
eigenfunction at the merging point of the singularities. In order to estimate the height
of this peak we set A = C, thus obtaining the estimate
Γ (d)Γ (c + n)Γ (d − a − b − 2n)Γ (n + a + b − d) −n
h0 = s (1 + O(s)) (4.19)
Γ (c)Γ (d − a − n)Γ (d − b − n)Γ (2n + a + b − d)
for the yet arbitrary coefficient h0 .
† Of course, the terms appearing in the studied problem differ from exponentially small terms in the
problems with turning points. However, here we mean simply that they have different origin and form
in comparison with conventional power-type asymptotics.

Proc. R. Soc. Lond. A (1999)

4356 W. Lay and S. Yu. Slavyanov

U
2
1

Figure 2. Dislocation in Peierls’ potential.

5. A physical example
Here, an example is studied with the help of which one may check our computation
in the case of a practically important physical model.
The plastic deformation of crystalline materials under a stress σ—caused by an
external force—is due to the migration of dislocations as one-dimensional atomic
defects. In the case of cubic space-centred crystals, the so-called ‘line-tension model’
(Lay 1991) is appropriate to describe adequately this migration process. The dis-
location in its crystalline environment is modelled by an infinitely long, stretched,
elastic string moving in a potential that has the shape of a sloped corrugated iron
sheet (see figure 2). The magnitude of the slope is according to the magnitude of the
stress generated by the external force. The migration of the dislocation takes place
in a periodic potential denoted U that is called Peierls’ potential and is caused by
the surrounding atoms. For modelling an elementary migration step, however, it is
sufficient to take a quartic (thus double-well) polynomial as U (called the Eshelby
potential). The direction of the dislocation migration is denoted by u. The elementary
migration step is from a straight (and thus stable static) configuration u = u(x) ≡ 0
in one well of the potential U (configuration (1) in figure 2) to a neighbouring lower
straight (and thus stable static) configuration u = u(x) ≡ 1 (configuration (3) in
figure 2) by going through a so-called kink-pair configuration ukp (x) (denoted as
configuration (2) in figure 2). A kink-pair is an unstable but static configuration
with lowest energy compared with all its ‘neighbouring’ configurations (e.g. a saddle
ridge configuration).
The dislocation migration rate in this model is governed mainly by the discrete
energy spectrum of this kink pair. It is determined by a Sturm–Liouville problem for
the one-dimensional Schrödinger type differential equation

d2 v
+ [E − V (x)]v(x) = 0, x∈R (5.1)
dx2

Proc. R. Soc. Lond. A (1999)

 Heun’s equation with nearby singularities 4357

V(x)

−x0 x0 x

−1

Figure 3. The double-well potential in Schrödinger’s equation.

having a potential of double-well form (see figure 3)

d2 U
V (x) = ∼ − sech2 (x + x0 ) − sech2 (x − x0 ). (5.2)
du2kp
The locations x = ±x0 of the two wells depend on the external force: if the stress
takes on its maximum value σ = σmax the quantity x0 has a minimal value that is
larger than zero. If the stress tends to zero σ → 0, then x0 → ∞.
Equation (5.1) is yielded by a linearization about the kink-pair configuration of
the wave equation that governs the dynamics of the string moving in the potential
U as described above and a subsequent plain wave separation ansatz.
It is clear from the behaviour of the potential (5.2) at infinity that the underlying
Schrödinger equation has only a finite number of discrete eigenvalues. Moreover,
because of the symmetry of V (x) with respect to x = 0 the eigenvalues group into
pairs, each of which consists of an eigenvalue belonging to an even and to an odd
eigensolution. The two levels of one such group approach each other when x0 increases
(this means a decrease of the stress due to the external force). It is known that this
approaching is an exponentially small effect with respect to the distance between the
two wells (Slavyanov 1996).
In his doctoral thesis, one of the authors (Lay 1987) has calculated the energy
spectrum of this system by converting equation (5.1) into a Heun equation (x →
ukp =: z, σ → s, v → y, E → 3) and using continued fraction methods (see Erdélyi
1955, p. 57). This Heun equation as the starting equation is given by

d2 y 1 2 1 1 dy 1
2
+ + + +
dz 2 z z − s z − 1 dz z(z − 1)(z − s)
32 s
× − − 6z + 3(1 + s) y(z) = 0, (5.3)
z
the RS of which is
 
0 s 1 ∞ ;z
3 0 0 −2 3(1 + s) . (5.4)
1 1
−3 2 2 3
The conversion from (5.1) into (5.3) is achieved by rewriting the Schrödinger equation
in such a form that the dependent variable of the function that describes the kink
pair becomes the independent variable of the differential equation (5.3). For the sake
of convenience this variable is no longer denoted by ukp but by z in (5.3), (5.4) and
in the following.

Proc. R. Soc. Lond. A (1999)

4358 W. Lay and S. Yu. Slavyanov

ρ
1.5

1.0

0.5

ξ
−1 0 +1

Figure 4. Eigenvalue curves of Heun’s equation.

It should be stressed that there will be the same eigenvalue curves ρ = ρ(s) when
s varies between 0 and 1 or between infinity and 1. This symmetry with respect to
the interval of s can be made obvious by using, instead of s, the new parameter ξ
according to
1−s 1−ξ
ξ= , s= .
1+s 1+ξ
The relevant physical interval on which the boundary-value problem has to be solved
is [0, s] for s < 1 (or, for the case stressed above, [s, ∞] for s > 1), where the
external force acting on the crystal determines the location s of the singularity of
the differential equation (5.3).
The actual calculation in Lay (1987) exhibited two pairs of eigenvalue curves for
equation (5.1). The numerical procedure breaks down when the singularity at z = s
merges with the singularity at z = 1. This manifests itself by the infinite continued
fraction becoming divergent. In our underlying physical model, s → 0 means the
maximum external stress above which the model becomes unstable and s → 1 means
the limiting case of vanishing external stress. The numerical results are shown in
figure 4.
As is revealed from the previous considerations, the reason for this breakdown is
the non-analyticity of the curves 3 = 3(s) at the merging point s = 1 or ξ = 0
of the two singularities. One of the authors called attention to this phenomenon
almost ten years ago (Lay 1987, 1990). Another example of this phenomenon occurs
in astrophysics on calculating the quasi-normal modes of rotating black holes (Leaver
1985).
In the following we exhibit a comparison of results between our asymptotics and
the numerical calculations that convinced us of the correctness of our theory. We

Proc. R. Soc. Lond. A (1999)

 Heun’s equation with nearby singularities 4359

start with an equation appropriate for odd eigensolutions in the following canonical
form:
3 3 
d2 w0 1 + 23 2 dw0
2
+ + + 2
dz z z − 1 z − s dz
(3 + 4)(3 − 1)z − (1 + s)(32 + 32 3 − 52 )
+ w0 = 0, (5.5)
z(z − 1)(z − s)
which results from equation (5.3) under substitution of y for w0 :

y = z ρ (z − s)1/2 (z − 1)1/2 w0 . (5.6)

In the notation we used in the former sections of this paper, equation (5.5) reads
3 3 
d2 w0 a+b−2 2 dw0 abz − λ
+ + + 2 + w0 = 0 (5.7)
dz 2 z z − 1 z − s dz z(z − 1)(z − s)
with
a = 3 + 4, b = 3 = 1, λ = (1 + s)(32 + 32 3 − 52 ) = ab − 34 (a + b) + 13
4 .

Equation (5.7) is a Heun equation by means of

c = 32 , d = a + b − 2, a, b, λ.
The problem studied in relation to equation (5.5) was the following: what are the
values of 3 at s = 1? The results obtained above can be straightforwardly carried
over and we obtain for the eigenvalues the result
λ = n(d − a − b − n) = −n(n + 2). (5.8)
Suppose that n = 0. Then, the following equation holds for 3 at s = 1:
232 + 33 − 5 = 0 (5.9)
having the solutions
3 = 1, 3 = − 52 . (5.10)
Only the first value is reasonable for the physical problem and has been found by
numerical computations (cf. figure 4).
Suppose now that n = 1. Then we have
232 + 33 − 2 = 0, (5.11)
the solutions of which are
3 = 12 , 3 = −2. (5.12)
Once again only the first solution is valid by physical reasons. For n
2 the corre-
sponding equations have complex solutions expressing the above-mentioned limiting
of the numbers of eigenvalues. The boundary problem for even eigensolutions of
equation (5.1) is studied in a similar way.
All facts discussed above were discovered numerically in Lay (1990).

Proc. R. Soc. Lond. A (1999)

4360 W. Lay and S. Yu. Slavyanov

6. Summary
The following conjectures may be formulated as a result of our studies. When two reg-
ular singularities in Heun’s differential equation merge, the corresponding boundary
problem is characterized by a boundary-layer phenomenon. Due to this phenomenon
the eigenvalues cannot be expanded in a power series of a small parameter (in our
case denoted by s). The reason is that they include terms which exhibit a branch-
point behaviour at the origin s = 0. These terms become smaller when the number
of the eigenvalues increases. A peak near the origin is typical for the corresponding
eigenfunctions. These qualitative features are valid if the conditions (1.3) hold. In
the limiting point s = 0 no explicit eigenfunctions exist although at this point the
eigenvalue curves have an explicit limit. The problem of two merging singularities is
closely related to that of two merging wells.
It is reasonable to assume that these conjectures are valid for a broader class
of equations with polynomial coefficients. However, the resulting formulae in these
more complicated cases are not explicit. In the neighbourhood of the merging regu-
lar singularities, solutions of the hypergeometric equation can once again be used as
solutions of a comparison equation as an Airy equation is used in the case of turning
points. The other auxiliary equation outside this neighbourhood (an equation with-
out a small parameter!) would, in the general case, be outside the hypergeometric
class of equations. This means that, presumably, the problem studied in this paper,
in Abramov & Slavyanov (1978) and, possibly, the problem of two merging wells, are
the only ones where the proposed procedure is valid as a whole. In other problems
the result is available in other terms.
The authors are pleased to thank Professor E. Kalnins for reading the article critically and K.-H.
Bay for numerical computations.
The research was partly sponsored (S.Y.S.) by RFFR grants 96-01-00-618, 98-01-01-075, and
by a grant of the Deutsche Forschungsgemeinschaft (Bonn).

References
Abramov, D. I. & Slavyanov, S. Yu. 1978 The two Coulomb centres problem at small inter-centre
separations. J. Phys. B 11, 2229–2241.
Bieberbach, L. 1965 Theorie der gewöhnlichen Differentialgleichungen. Berlin: Springer.
Erdélyi, A. (ed.) 1995 Higher transcendental functions, vol. 3. Bateman Manuscript Project.
New York: McGraw-Hill.
Fedorjuk, M. V. 1993 Asymptotic analysis: linear ordinary differential equations. New York:
Springer.
Lay, W. 1987 Ein Beitrag zur mathematischen Behandlung von Kinkpaaren in Versetzungen.
Dissertation thesis, Stuttgart.
Lay, W. 1990 Symmetries in eigenvalue problems of Heun’s equation in the case of a nontrivial
confluence. In Centennial Workshop on Heun’s Equation: Theory and Applications (ed. A.
Seeger & W. Lay). Max-Planck-Institut für Metallforschung.
Lay, W. 1991 Kink pair formation on dislocations within the framework of the line tension model.
In Proc. Adriatico Research Conf. on Quantum Fluctuations in Mesoscopic and Macroscopic
Systems, Miramare, Trieste, 3–6 July 1990 (ed. H. A. Cerdeira, F. G. López & U. Weiss),
pp. 97–120. Singapore: World Scientific.
Leaver, E. W. 1985 An analytic representation for the quasi-normal modes of Kerr black holes.
Proc. R. Soc. Lond. A 402, 285–298.

Proc. R. Soc. Lond. A (1999)

 Heun’s equation with nearby singularities 4361

Olver, F. W. 1997 Asymptotics and special functions. Wellesley, MA: AKP.

Seeger, A., Lay, W. & Slavyanov, S. 1995 Confluence of Fuchsian second-order differential equa-
tions. Theor. Math. Phys. 104, 950–960.
Slavyanov, S. Yu. 1996 Asymptotic solutions of the one-dimensional Schrödinger equation. Trans.
of Mathematical Monographs, vol. 151. Providence, RI: AMS.
Slavyanov, S. Yu., Lay, W. & Seeger, A. 1997 A generalization of the Riemann scheme for
equations of the hypergeometric and Heun classes. Int. Jl Math. Educ. Sci. Technol. 28,
641–660.

Proc. R. Soc. Lond. A (1999)