Theoretical and Mathematical Physics, Vol. 104, No.

2, 1995

C O N F L U E N C E OF F U C H S I A N SECOND-ORDER
DIFFERENTIAL EQUATIONS
A. Seeger, W. Lay, and S. Yu. Slavyanov

Ordinary linear homogeneous second-order differential equations with polynomial coefficients including one in
front of the second derivative are. studied. Fundamental definitions for these equations: of s-rank of the singu-
larity (different from Poincarg rank), of s-multisymbol of the equation, and of s-homotopic transformations are
proposed. The generalization of Fuchs ' theorem for confluent Fuchsian equations is proved. The tree structure
of types of equations is shown, and the generalized confluence theorem is proved.

1. Introduction
Linear homogeneous second-order equations with polynomial coefficients including that of the second
derivative (the so-called Fuehsian and confluent Fuchsian second-order equations) and their solutions, some
of which are special functions, form an important part of the branch of mathematical physics that describes
properties of the simplest models. The literature on special functions is used regularly by the majority of
theoretical physicists. However, it turns out to be rather difficult to find information about the solutions
of more complicated equations in the standard reference books. The purpose of this paper is to point
out certain regularities of the sets of such equations and to describe the convenient tree-like structure
of the types of equations that would help in studying properties of the solutions and in making possible
simplifications.
We used ideas of the classical works [1-4] as the starting point in our research, although a considerable
part of them was completely revised. The notion of s-rank of the singularity is basic for the classification,
and the main theorems are proved using this notion: the generalization of the Fuchs theorem for the
confluent equations and the theorem concerning subadditive property of s-rank under a confluence. In
our opinion the classification of parameters and the classification of some stazldard forms of the equations
deserve special attention as well. According to this classification, special functions are grouped into certain
classes. The first one consists of hypergeometric functions, and the second one is the Heun class.
In this paper we present the proofs of new results, only not the classical ones. All the notions are
introduced in the form of definitions.
2. Second-order Fuchsian equations
Our starting point is an ordinary linear homogeneous second-order differential equation

Po(z)y"(z) + P l ( Z ) y ' ( z ) + P~(~)y(z) = 0, z e c, (1)

with polynomial coefficients Po(z), Pl(z), and P2(z), Their degrees are no, nl, and n2 respectively. We
also assume that these polynomials are relatively prime.
The singularities of Eq. (1) are zeros of the polynomial Po(z), i.e., points zj (j = 1 , . . . , n) and probably
Z ~ (X3.

Equation (1) is called Fuchsian if all its singularities are regular. The equation remains Fuchsian and
the number of its singularities does not change under isomorphisms C of the z-plane that establish the
equivalence relation on the set of the Fuchsian equations. Thus, we can assume, without loss of generality,
that the point z = oc is a regular singularity for all of the equations considered. The set of all singularities

St.-Petersburg State University. Translated from Teoreticheskaya i Matematicheskaya Fizika, Vol. 104, No. 2,
pp. 233-247, August, 1995. Original article submitted October 25, 1994.

950 0040-5779/95/1042-0950512.50 ©1996 Plenum Publishing Corporation

of Eq. (1) will be denoted as {zj; oe}, j = 1,..., n. We also assume that n > 1 in order to exclude possible
discrepancies.
Let us consider the dependent variable transformation of Eq. (1) of the form
s: y w, v(z) = w ( z ) v ( z ) , (2)
where v(z) is a solution of the following auxiliary first-order equation with polynomial coefficients:
Qo(z)v'(z) + Ql(z)v(z) = 0. (3)
Assume that the set of zeros of the polynomial Qo(z) coincides with {yj} and can be written in the form
/2

Qo(z) = I I ( z - zj);
j=l

Q(z) is an arbitrary polynomial with degree not exceeding n - 1.

D e f i n i t i o n 1. Equation (3) is called associated to Eq. (1) under these assumptions.
The function

,(z) = H ( z - zj)"J
j=l

with arbitrary #j (some of them may be zeros) is the solution of Eq. (3). The transformation (2) converts
Eq. (1) into a new equation

Pow"(z)+ ( P 1 - 2 P o Q ~ ) w ' ( z ) + (P2+ (Q~ +Q~oQI

Q2 _~o)Po _ ~00Pl)
Q1 w(z) -- 0. (4)

D e f i n i t i o n 2. A Fuchsian equation is called irreducible if under any transformation of the form (2)
the number of its singularities [from Eq. (1) to Eq. (4)] is not reduced. Otherwise, it is called reducible.
Similarly, we can discuss reducible and irreducible regular singularities. A transformation of the form (2)
establishes another equivalence relation on the set of all irreducible Fuchsian equations.
D e f i n i t i o n 3. Any equation equivalent to (1) is called a form of Eq. (1). The form of the Fuchsian
equation is called canonical if every zero of the polynomial Po(z) is prime. The form of the Fuchsian
equation is called natural if z = cx~ is its regular point.
It was already mentioned that any Fuchsian equation can be transformed into its natural form by an
isomorphism of the z-plane (such mapping is not unique).
L e m m a 1. For any irreducible Fuchsian equation in its natural form there exists a transformation of
the form (2) which converts the equation into the canonical form. This transformation is not unique.
P r o o f . We construct this transformation by choosing #j as any soluiion of the equation
p(p-1)+pjp+qi=O, p=pl(zj), p=p2(zj), (5)
determined by the point zj, where
pj = Res Pl(z)/Po(z), qj = z=zj
Res(z - zj)P2(z)/Po(z). (6)
z=zj
The quantities pro(zj) (m ---- 1,2) are called the characteristic indices at the point zj. The characteristic
indices at infinity pro(co) (m = I, 2) are defined as solutions of the equation
p(p+l) Tp~p+qc~=O, p=pa(oo), p=p2(oo), (7)
where
poo = - Res Pl(z)/Po(z), qj = Res P2(z)/Po(z). (S)

These indices characterize nonholomorphic components of the Frobenius solutions of Eq. (1) in the neigh-
borhoods of the regular singularities.

951
 T h e o r e m 1 (Fuchs' theorem, [5]). The equality

2 2

Pm(ZJ) -1- ~ Pro(00) = n - 1 (9)

j=l m = l m=l

holds for the sum of all characteristic indices of the Fuchsian equation.
The subset of elementary singularities of the set of regular singularities of the Fuchsian equation is of
special interest [1]. A regular singularity zk is called elementary if the difference of characteristic indices at
this point is equal to one-halL There exist a transformation of the form (2) and a quadratic transformation
of the independent variable T : z ~+ t, z - zk = t 2 which remove this elementary singularity acting together.
Definition 4. The parametric adjustment of polynomial coefficients of P1 and P2 which converts the
general regular singularity into an elementary one is called the reduction of the regular singularity.
Definition 5. The Fuchsian equation with partly reduced singularities is called the reduced Fuchsian
equation.
E x a m p l e s . The Riemann equation is the Fuchsian equation with three singularities; the hypergeometric
equation is its natural canonical form. The equation with one reduced singularity can be transformed
into the associated Legendre equation; the equation with two reduced singularities - - into the equation
generating Chebyshev polynomials.

3. Confluent fuchsian equations

Definition 6. The Fuchsian equation (1) is called confluent if it has at least one irregular singularity
besides the regular ones.
We assume further that infinity is one of the singularities, as has been assumed for the Fuchsian equations.
We shall specify every singularity by the numerical value called s-rank of the singularity. This differs
from the value of the singularity rank introduced by Poincard.
Definition 7. The s-rank Rj of the finite irregular singularity zj is

R(zj) = max(K1 (z j), I<2(zj)/2), (i0)

where K1 (zj) is the multiplicity of zero of the function Po(z)/Pl(Z) at zj and Ke(zj) is the multiplicity of
zero of the function Po(z)/P2(z) at the same point. The s-rank R(oo) at the infinite irregular point z = oo
is
R(oo) = max(K1 (oo), K2(oo)/2), (11)
where K,(oo) = n, - no + 2 and lg2(oo) = n2 - no + 4.
The S-rank of an irregular point is an integer or half-integer exceeding one.
Definition 8. We define the S-rank of a regular nonetementary singularity equal to one. This agrees
with Definition 7.
The s-rank of an elementary singularity can be defined in various ways. There are several reasons to set
it equal to one as is defined for other regular singularities. According to historical traditions we decided to
define it separately to show that elementary singularities play the special role emphasized in many classical
works.
Definition 9. The S-rank of an elementary singularity is equal to one-halE
Definition 10. An irregular singularity with half-integer s-rank is called a reduced irregular singularity.
If the s-rank is an integer, the singularity is called nonreduced. The regular nonelementary singularities are
also nonreduced.
Definition 11. The set of s-ranks of all singularities of the confluent Fuchsian equation forms the
s-multisymboI of the equation.

952
 E x a m p l e s . The s-multisymbol of the confluent hypergeometric equation is {1, 2}. The s-multisymbol
of the equation generating Hermitian polynomials is {3}. The s-multisymbol of the Airy equation is {5/2}.
Now we shall present a generalized transformation of the form (2) for the confluent Fuchsian equation
(1). Its construction is also based on the associated first-order equation (3); however, the polynomials Q0
and Q1 have to satisfy conditions different from those above.
D e f i n i t i o n 12. The transformation of Eq. (1) defined by formulas (2), (3) is called s-hornotopic if:
1) the set of zeros of Qo(z) coincides with {zj};
2) the multiplicity r(zj) of any zero of the polynomial zj (j = 1,..., n) does not exceed the a-rank of
zj (j = 1,... ,n) as the singularity of Eq. (1);
3) the difference of degrees of Qo(z) and Q~ (z) (10, 11 respectively) satisfies the following inequality:
ll --lo ~ max(-rX'l ( o c ) , / ( 2 ( o c ) / 2 ) -- 2.

We can represent the fraction Q1/Qo as the sum of the simplest terms:

Ql(z) ,~ ,-(~j) t,-to

E E H i k ( z - zi) -k + E Fkzk" (12)
Qo(z)
-

j=l k=l k=O

From (t2) we can get the explicit expression for the function v(z) acting in s-homotopic transformation of
the form (2):

Qldz~ = e x p Fkl--~ (z - - )gJ~ exp Yik

v(z) = exp -- Q0 /} ', k=0 = \k=2

As a result of the s-homotopic transformation of the input equation the s-rank of any singularity can either
remain the same or decrease.
D e f i n i t i o n 13. The s-rank singularity is called reducible if there exists an s-homotopic transformation
which decreases the s-rank of the singularity. Otherwise, it is called irreducible.
D e f i n i t i o n 14. If all of the singularities of the confluent Fuchsian equation are irreducible, then the
equation is also called irreducible. Otherwise it is called reducible.
S-homotopic transformations establish the equivalence relation on the set of irreducible equations. Iso-
morphisms of the complex plane of independent variables establish another equivalence relation on the same
set.
D e f i n i t i o n 15. The confluent Fuchsian equations equivalent to the input equation are called the forms
of the input equation.
Later we shall introduce the canonical natural form of the confluent Fuchsian equation as has been
defined for the Fuchsian equation.
Hereafter we assume that every singularity of the Fuchsian equation and the Fuchsian equations are
irreducible, unless otherwise stipulated. A local behavior of the solution in the neighborhood of irregular
singuIaxity may be characterized by its formal asymptotic expansion (the Tome expansions) [6]. If the point
is finite and nonreduced, these series are in the form

gin(z) = (z - zj)~"°(ZJ) exp - zj) -k E cmk(zj)(z - zj) k, rn = 1,2. (14)

\ k=, (-k) k=0

The series of the form (14) axe called normal solutions. If the point is finite and reduced the Tome expansions
take the form

ym(Z) = -- z j ) exp ( z -- z j ) Cm (Zj)(Z -- m = 1,2, (15)

\ k= /2 (-k)

953
where the index k in the exponential takes integer as well half-integer values. The series of the form (15)
are called subnormal solutions. If the singularity is at the point z = oo we have to replace expansions (14)
by

ym(z) = z -~'~°(°°) exp \

(R(~__~--I
k_-i
Otmk(~ ) k~ oo
z / m = 1,2, (16)

and expansion (15) by

ym(z) = z -~m°(°¢) exp R( 1 am--OO)zk Ecm'(°°)z-'/2' rn = 1,2, (17)

\ k=1/2 k=0
D e f i n i t i o n 16. The multipliers of the sums in (14)-(17) are called the determining multipliers of the
Tome solutions.
D e f i n i t i o n 17. The coefficients amk in the expansions (14), (t6) are called the second-kind characteristic
indices of order k of a nonreduced irregular singularity. They form two characteristic sequences. It is
possible to evaluate the terms of these sequences by the coefficients of the differential equation using certain
recursion procedures.
D e f i n i t i o n 18. The coefficients in the expansions (15), (17) are called the second-kind characteristic
indices of order k of a reduced irregular singularity. They form two characteristic sequences. It is possible to
evaluate the terms of these sequences by the coefficients of the differential equation using certain recursion
procedures.
L e r n m a 2. The characteristic indices of the reduced irregular singularity satisfy the following conditions:
1) the equality
alk = a2k (18)
holds for the characteristic indices of the integer order;
2) the equality
alk = --C~2k (19)
holds for the characteristic indices of the half-integer order.
P r o o f . Let us consider only the finite, reduced and irregular singularity zj. Let us represent the deter-
mining multiplier at this point in the form

where the function qSl(Z) includes only integer characteristic indices and q52(z) includes only half-integer
characteristic indices. Then we shall substitute the Tome solution into Eq. (1) and isolate the highest
singular terms. The sequential reduction of these terms will lead us to amk(zj). The separation of the
terms with integer powers of z - zj from the terms with half-integer powers leads us to the equations
¢2 + ¢2 + P~/Po + e~ = 0, (20)
2~b2~b1 + ~b2Pa/(2P0) + ~b; + ~1 = 0, (21)
where el(z) and 5~(z) already do not have the highest singularities at z - zj. The 1emma follows from (20)
and (21).
L e m m a 3. The reduced singularities are nonreducibte.
P r o o f . The temma assertion follows from a comparison of the determining multipliers of the Tome
solutions which contain half-integer powers under the exponential with the function v(z) in s-homotopic
transformation which contains only integer powers under the exponential.
Now we shall construct the canonical form of the confluent Fuchsian equation using the formulas (14)-
(17). For example,the canonical form was used in [7] in the description of the confluent Heun's equations.
Let us introduce the chain of preliminary definitions.

954
 D e f i n i t i o n 19. The form of Eq. (1) at a finite nonreduced irregular singularity zj is called canonical if
one of the sequences of the characteristic indices of the Tome solutions is a zero sequence in the neighborhood
of zj.
D e f i n i t i o n 20. The form of Eq. (1) at an infinite nonreduced irregular singularity z = oc is called
canonical if there exists Tome solutions such that the second-genus characteristic indices of positive order
are zeros.

D e f i n i t i o n 21. The form of Eq. (1) at a finite reduced singularity is called canonical if both Tome
solutions at this point have zero second-genus characteristic indices of positive integer orders, and its
second-genus characteristic indices of zero order differ only by their signs.
D e f i n i t i o n 22. The form of Eq. (1) at a finite reduced singularity zj is called canonical if both Tome
solutions at this point have zero second-genus characteristic indices of positive integer orders.
D e f i n i t i o n 23. The form of Eq. (1) at a finite regular singularity is called canonical if one of the
characteristic indices at this point is zero.
D e f i n i t i o n 24. The form of Eq. (1) at a finite elementary singularity is called canonical if one of the
characteristic indices at this point is zero.
D e f i n i t i o n 25. The form of the confluent Fuchsian equation is called natural if the infinite point is the
singularity with the maximal s-rank.
D e f i n i t i o n 26. The form of the confluent Fuchsian equation is called natural canonical if it is canonical
at every singularity and the infinite point is the singularity with the maximal s-rank.
L e m m a 4. The equation is canonical at nonreduced finite singularity zj if and only if the zero multiplicity
of the function Po(z)/P1 (z) at zj is equal to its s-rank and the zero multiplicity of the function Po(z)/P~(z)
does not exceed its s-rank.
L e m m a 5. The equation is canonical at the reduced finite singularity zj if and only if the function
Po(z)/Pl(z) does not vanish at this point,
P r o o f . The 1emma assertion follows from Eq. (21) for the characteristic indexes of integer order. They
are zeros if the point zj is not a singular one for the function P1/19o. On the contrary, if they are zeros,
then the point cannot be singular.
T h e o r e m 2. Any confluent irreducible Fuchsian equation can be transformed into its canonical natural
form by the superposition of an isomorphism of the complex plane of independent variables and s-homotopic
transformations.
P r o o f . How to transform the equation into its natural form is quite obvious. The property of the
form to be canonical at elementary and regular singularities can be obtained by the method of Lemma 1.
So, let us consider the finite nonreduced singularity. The function v(z) which performs the s-homotopic
transformation into the form canonical at this point is chosen to be equal to the determining multiplier of
one of the Tome solutions at this point. None of the characteristic indices of the other singularities except
the zero order index of the infinite point will be changed under this operation. Let us consider a finite
reduced singularity zj. At this point the function v(z) which performs the s-homotopic transformation into
the form canonical at this point is chosen as

.(z) = (z - ) exp (R(~3/2alk(zj)(z_zj)_k)

\ (-k)"

where alk(Zj) are integer order characteristic indices at this point. If the singularity is infinite we should
exclude the multiplier (z - zj)~l°(zD from this formula.
Let us introduce the notion of normal natural form of the confluent Fuchsian equation.
D e f i n i t i o n 27. The form of the confluent Fuchsian equation is called normal natural, if its coefficient
P1 (z) = 0 and the infinite point is the maximal s-rank singularity of this equation.

955
 T h e o r e m 3. Any confluent irreducible Fuchsian equation can be translormed into its normal natural
form by the superposition of isomorphism of the complex plane of independent variable and s-homotopic
transformations.
P r o o f . The transformation into the natural form is obvious. The transformation into the canonical
form is well known.
Let us state without proof that any confluent Fuchsian equation is irreducible in its canonical natural
form and in its normal natural form.
4. G e n e r a l i z a t i o n s o f t h e F u c h s t h e o r e m
In this section we prove several generalizations of the Fuchs theorem using the notions introduced in
Section 3.
T h e o r e m 4. Let all of the singularities of the confluent Fuchsian equation (1) be neither reduced nor
elementary. Let n~ denote the number of regular singularities and ni the number of irregular singularities.
Then the identity
nr 2 ni 2 2 ni'J-nr
E E pm(Zj)-t- E E OlraO(ZJ)~- E Otm0(OO)----" E .R(zj)"~- i~(oo)- 2 (22)
j = l m=l j = I rn=l m=l j=t
holds. All of the characteristic indices of zero order of Frobenius and Tome solutions are summed up on
the left-hand side of this identity; all of the s-ranks of the equation singularities are summed up on its
right-hand side.
P r o o f . First of all it should be mentioned that s-homotopic transformations change neither the left-
hand side nor the right-hand side of this identity. Therefore, we can consider the canonical natural form of
the equation. Any function f(z) with a pole at the point zj can be represented as a sum of the principal
Gj and the regular "r.f." parts of the Laurent series at this point. In particular,
R(zj)
P1/Po = Gj + r . f , Gj = E a_~-(zj)(z - zj) -k. (23)
k=l
Let Tj be the logarithmic derivative of the determining multiplier at the point zj. It is equal to
R(~j)-I
Tj = amO(Zj)(z - zj) -1 + E a,~k(Zj)(z -- zj) -k-1. (24)
k=l
Let us substitute the Tome solution with the determining multiplier not equal to one into Eq. (1) and
separate out the terms with the highest singularities. Thus we obtain the recursive equation for nonzero
sequence of the second-genus characteristic indices:
T] + GjTj + T; = O, (25)
where C~lk should be chosen such that they, annihilate all singularities of order from - 2 R ( z j ) to - R ( z j ) - 1
inclusively. The recursive procedure of finding a1 k (zj) leads to
= -a_k(zj), k = 2,..., R(zj), (26)
,o(zj) = R ( z , ) - a-l(z,). (27)
The form of the equation is canonical, so all c~2k(zj) = 0. We can examine all the finite irregular singularities
in the same manner. Now, let us consider the infinite irregular point. In its neighborhood the representation
(23) must be changed to
~c¢~12
pl/Po = + O(z-2), = bkz k ,
k=--I
P2/PI = c(oo)z -2 + O(z-a), (28)

956
and Eq. (25) to
c(oo)
T} + CooTj + T ; + - -z2- - 0. (29)

The characteristic indices of the zero order are equal to

c(co) + (R(oo)--2) +~a-l(zj),

420(00) -- bR(oo) - - 2
j=l
c(co) (30)
alo(co) = bn(~) - 2"

Adding (27) to (30) we obtain (20).

If we map C isomorphically and recount both sides of the identity (22) to match this isomorphism we
can verify that the initial choice of the natural form is significant only for the exterior of the identity but
not for its entity.
Now let us assume the existence of reduced singularities of Eq. (1).
T h e o r e m 5. Suppose that the confluent irreduced Fuchsian equation does not have elementary sin-
gularities; it has n~ regular singularities, ni irregular nonreduced singularities, and ni~ irregular reduced
singularities. The form of the equation is natural. Then the identity (22) holds if we replace ni by ni + ni~
in the sum.
P r o o f . Let us examine a finite reduced singularity zj. First of all, we reduce the equation to its canonical
form at this point. This means that this point is not a singular point for the function P2/P1 and Eq. (21)
serves only for evaluation of the characteristic indices C~mo(zj). After calculations we get

mO(Zj) - m z2j ) ' m = 1,2. (31)

This formula shows that the assumption of the existence of the finite reduced singularities of Eq. (1) does
not change the formula (22). Now let us examine what happens if the point z -- co is also a reduced
singularity. We take the equation in its canonical form. Equation (21) gives the following formula for the
coefficients C~m0(co):
nr+nl
20~mO(OO) = (.~(OO) -- 2) + Z a-l(Zj), ~2 = 1,2. (32)
j=l

This formula implies that (22) is valid in this case as well.

Finally, we consider the equation with elementary singularities.
T h e o r e m 6. Let us assume that the confluent Fuchsian equation has elementary, regular nonelementary,
irregular reduced, and regular nonreduced singularities. Their numbers are respectively he, nr, hi, and ni~.
The form of the equation is natural. Then the identity

nr+ne 2 niWnir 2 2 nr+ni+nir

(33)
j=l ra=l j=l rn=I rn=l j=l

holds.
P r o o f . The proof repeats t h a t of Theorems 4 and 5.
One can notice that our definition of the rank of an elementary singularity leads to certain violations in
the formula (33). It would have been more natural to have the sum of all s-ranks on the right-hand side of
the identity (33).

957
5. T h e c o n f l u e n c e a n d t h e r e d u c t i o n o f t h e e q u a t i o n
It has been shown by Ince [1] that every singularity can be conceived as the result of the confluence (or
several confluences) of elementary singularities. Since then, the classification of the types of equations has
been based on the confluence of singularities. In our opinion, however, taking into consideration the process
of confluence only will lead to a one-sided approach to the problem and will not allow one to define the
classes of special functions with similar properties. Only parallel usage of both confluence and reduction
can provide such an opportunity.
D e f i n i t i o n 28. In a broad sense the word confluence means the confluence of two singularities of Eq.
(1) into one with the s-rank exceeding the s-ranks of the input singularities under a parametric change of
the coefficients of the polynomials P1 and P2 with respect to the distance between them.
Now we should make several remarks: the intermediate s-homotopic transformations are admitted; the
new reducible singularities can appear under the confluence, but they are easily removed by s-hornotopic
transformations.
Moreover, we should not forget about the number of parameters. It is suggested that the input equation
has the maximal number of free parameters that are compatible with the s-multisymbol and the form of
the equation.
D e f i n i t i o n 29. The confluence is called strong if the number of free parameters decreases by one under
this confluence. If it decreases by more than one, the confluence is called weak.
We can avoid this complications by preliminary "preparation" of the confluent equation, namely, by
transforming it to a form in which the confluence is easily observed.
Using this simplification we prove the theorem about the additivity property of the s-rank under the
strong confluence (if the confluence is weak, then it is the subadditivity property).
T h e o r e m 7. Let a confluent Fuchsian equation in its normal form with one singularity zl = 0 of s-rank
R~ and another singularity zl = s of s-rank R2 be given. The process of strong confluence generated by the
limiting transition c --+ 0 leads to one resulting singularity z~ = 0 of s-rank 1~ = R1 + R2.
P r o o f . Let us consider the principal parts of the Laurent series of the function P2/Po in the neighbor-
hoods of both singularities:

2R1 2R2
P2(z)/Po(z) = Gl(z) + G2(z) + r.f., G,=Ea (34)
k=l k=l

We can convert the sum GI + G2 into

2R1 2R2
akz2Rl-k(z _ ¢)2R2 ÷ ~ bk(z --¢)2R2-kz2R'
G1 + G2 = k=l (35)
z~R1 (z - k=l
e)2R2

If all of the coefficients ak, bk are of order one, then in the limit e --+ 0 we should get a singularity of s-rank
equal to the maximum of the s-ranks of the input singularities. It is necessary ibr the confluence that these
coefficients grow as certain powers of e -1. The form of the denominator in (35) shows that the s-rank of
the resulting singularity can not exceed Re. Thus, the proof of the theorem is reduced to justification of
the possibility to choose the coefficients ak, bk depending on e to perform the strong confluence. We test
this choice by the following evaluations. Let the nominator of (34) tend to tile polynomial

2(RI+R2)
Gc = E f2(R'+R2)-kZ-k (36)
k=l

958
Comparing the coefficients of powers of z we obtain a system of equations. It is convenient to split it in
two subsystems: the first one is to find ak as polynomials in powers of e - l :

a2R1-1 (_¢)2R2 + f/1 (a2R1, e -1 ) = f2R2+2R1-1,

(37)

+ fhR,-l(a2R,,... - 1 ) = f2R +l,

and the second one to find bk:

bl + f~2Rl(a2R1,.--,a1,6 -1) = f l ,
b2 + f l 2 R l + l ( a 2 R 1 , . . . , a l , b l , g -1) = f2,
(38)

b2R2 + f~:R~-l(a2R~,...,al,bl,b2R~-l,e -1) = f2R2,

where the functions f~k are linear functions in the first arguments and polynomials in the last one. Actually.
we should have put "limits" on the left-hand sides of these equations, but we omit them to simplify the
arguments. We choose ak and bk recursively according to Eqs. (37), (38) as linear functions of the free
parameters fk in the limit c -+ 0• Thus, we have built the strong confluence of Eq. (1).
Definition 30. The reduction of a singularity is the special choice of the coefficients of polynomials Pl
and P2 which decreases its s-rank by half and converts a nonreduced irregular point into a reduced one,
and a nonelementary regular point into an elementary one.
However, not every singularity can be reduced, For example, it is impossible to reduce irregular sin-
gularity if the form of the equation is canonical, Thus, auxiliary s-homotopic transformations should be
involved in the process of reduction.
Now we shall define the classes of second-order equations with polynomial coefficients using both con-
fluence and reduction.
Definition 31. A nonreducible Fuchsian equation with n + 1 singularities and with arbitrary coefficients
of the polynomials P1 and P2 compatible with a chosen form of the equation is called a generatin 9 equation
of class M,~ of the second-order equations with polynomial coefficients.
Definition 32. The equations that are equivalent to the generating equation which appears under
confluence and reduction, and those that are equivalent to these new equations, form the class of second-
order equations with polynomial coefficients.
E x a m p l e s . The equation y" = 0 is generating for the class M0, and the Euler equation is generating for
the class M1. Both classes have elementary solutions only, so one should make some additional stipulations
to describe these classes. The hypergeometric equation is generating for the hypergeometric class -~I2. The
Heun's equation is generating for the Heun's class M3.
Definition 33. Definitions of the equations are distinguished within the classes. The type of the equation
is compIetely characterized by its s-multisymbol.
E x a m p l e s . There are 10 types of equations in the hypergeometric class. Their s-multisymbols are
depicted above. The Heun's class consists of 20 types of equations.

6. C o n c l u s i o n
Some of the definitions introduced in this paper may be the subject of discussion• The definition of the
s-rank of an elementary singularity equal to one-half is especially controversial. It is possible to define it
according to the formulas (10), (11) so that the difference of the characteristic indices would be equal to
one. However, this definition will not change the confluence theorem and will not help with the statement
of Theorem 6. Our choice of the definition was due primarily to some historical reasons as we wanted to

959
 " I
.%
\
\
\
\ ¢/z z .>
"t I z
"\\ I ;/z
~ 3/z i
\
\
N 5/, I

emphasize the peculiarity of elementary singularities and to obtain the well-known classification of special
functions.
This paper may be regarded as a basic paper. We present here the main definitions and the proofs of
the main theorems, A number of papers that examine this problem within the fl'amework of the presented
classification of some properties of several classes has been published already.
This paper is the joint work of three authors, but the text was written by S. Yu. Slavyanov, who is
responsible for the material.
The authors are grateful to Prof. F. Arscott and to Prof. D. Schmidt for their valuable comments. This
work was supported by the National Science Foundation, grant P5X000.
REFERENCES
1. E. L. Ince, Ordinary Differential Equations, Dover, New York (1959).
2. P. Moon and D. E. Spencer, Field Theory Handbook, Springer-Verlag, Berlin, Heidelberg, New York
(1971).
3. A. R. Forsyth, Theory of Differential Equations, Dover, New York (1959).
4. E. T. Whittaker and G. N. Watson, A Course of Modern Analysis, University Press, Cambridge (1963).
5. L. J. Fuchs, Reine Angew. Math., 66, 121 160 (1866).
6. L. Bieberbach, Theorie der gewShnlichen Differentialgleichungen, Springer-Verlag, Berlin, G6ttingen,
Heidelberg, New York (1965).
7. A. Decarreau, M. C. Dumont-Lepage, P. Maroni, A. Robert, and A. Ronveaux, Ann. Soc. Sci. Bruzelle,
92, 53-78 (1978).
8. W. Lay, A. Seeger, and S. Yu. Slavyanov, A Classification Scheme of Heun's Differential Equation and
Its Confluent Cases (in press).

960