MATHEMATICS OF COMPUTATION, VOLUME 26, NUMBER 117, JANUARY, 1972
Product Integration for the Generalized Abel Equation
By Richard Weiss
Abstract. The solution of the generalized Abel integral equation
gi') = [kit, *)/« - s)"}f(s) ds, 0 < a < 1,
where kit, s) is continuous, by the product integration analogue of the trapezoidal method
is examined. It is shown that this method has order two convergence for a £ [at. 1) with
ai = 0.2117. This interval contains the important case a = \. Convergence of order two
for a £ (0, ai) is discussed and illustrated numerically. The possibility of constructing
higher order methods is illustrated with an example.
1. Introduction. The generalized Abel integral equation
(1.1) git) = f S\ f(s)ds, 0<«<l,0g(gr<«=,Jo {t — s)
where k(t, s) is continuous for 0 g s rg t ^ T and
(1.2) k{t, <)^0,
is of considerable importance in many fields. For example, consider the equation
(1.3) Y(y) = 2 f1 2 R(r\1/2 dr, 0 g ^ 1,K 0- — y )
which occurs frequently in mathematical physics. By a change of variables, (1.3)
can be transformed to
(1.4) git) = f . Ks\l/2ds, O^t^l,J0 (t — i)
which is an important special case of (1.1). Various numerical methods for per-
forming the inversion of (1.3) as well as (1.4) have been developed.-A bibliography
can be found in Minerbo and Levy [1].
As a second example, consider equations of the type
Or)-172 exp[-Ma(0}7']
(1.5)
= / {2x(f - SV1'2} exp[-i{a(0 - ais)}2/« - *)]/(*) ds,
where a(t) is a differentiable function with o(0) > 0. Such equations and systems of
Received May 10, 1971.
AMS 1970 subject classifications. Primary 65L05.Key words and phrases. Generalized Abel equation, numerical solution, product integration
methods.
Copyright © 1972, American Mathematical Society
177
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
178 richard weiss
similar structure appear in the analysis of diffusion processes [see Fortet [2]}. A
method for solving (1.5) has been proposed by Durbin [3].
However, none of the methods suggested for (1.4) as well as (1.5) can be applied
to the general case (1.1). Only methods based on the concept of product integration
{see Young [4]} appear to be applicable. Such methods have been suggested on
the basis of rigorous numerical experimentation by Linz [5] and Noble [6]. Inde-
pendent support for this can be found in [3], though the product integration type
method developed there is not generally applicable.
A theoretical justification for the use of product integration methods can be
found in Weiss and Anderssen [7]. They show that the midpoint product integration
method is convergent of order one if f(s) and k(t, s) are suitably smooth and con-
vergent of order at least (1 — a) if weaker smoothness conditions apply.
In this paper, we extend the result of [7] by examining the applicability of other
product integration methods to (1.1). Initially, we show that the use of trapezoidal
product integration leads to a method with second order convergence. The result
again depends on the smoothness of /(r) and k(t, s). In addition, the possibility of
determining higher order methods is discussed.
The existence and uniqueness of solutions of (1.1) is considered in Section 2.
In Section 3, we give a basic definition and a lemma required in the subsequent
analysis of Section 4, where we define the trapezoidal method and investigate its
convergence. In Section 5, we examine the effect of computational errors, and a
numerical example is presented in Section 6. A third order method is considered
in Section 7.
2. Analytic Solution of (1.1). The existence and uniqueness of solutions of
(1.1) is guaranteed under the following conditions:
Theorem 2.1 {Kowalewski [8, Section 1, pp. 80-82]}. The integral equation
(1.1) has a unique continuous solution /(r), t £ [0, T], if k(t, s) and g(f) satisfy
(i) k(t, s) and dk(t, s)/dt are continuous for 0^1^(51,
(ii) k(t, t) * Ofort £ [0, T],(iii) G(r) = d/dt f'Q (g(s)/(t - s)1'") ds is continuous for t £ [0, T].
3. Preliminaries. To solve (1.1) we introduce the grid {r, - /Ar, / = 0, 1, • • • ,
/ = T/At} with gridspacing At. An approximate solution, defined with respect to
these grid points, will be denoted by /,.
Definition 3.1. Let Ji(At) (i = 0, 1, ■ • •) denote the approximation to /(r,) ob-
tained by a finite difference method with gridspacing At. Then, the method is said
to be convergent if and only if
max 17, (AO - fitdl -* 0OSiS/
as Ar —> 0, I —> oo such that I At = T. Furthermore, the method is said to be con-
vergent of order p, if p is the largest value for which there exists a finite constant C
and a Ar0 such that
max I/,(Ar) - /(r,)l ^ CAr"0S.S7
for all 0 < At ^ At0.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
the generalized abel integral equation 179
The convergence proof developed in Section 4 is based on the following lemma
which is a consequence of the standard results for regular infinite systems of algebraic
equations {see Kantorovich and Krylov [9, Section 2, p. 27]}.
Lemma 3.1. Let the real numbers xtii — 1» 2, • • •) satisfy
|*iI ^ K, \x2\ ?s K, K is const,
l**+»l =s E k.,i \xA + \b<\ a = 2,3, •••),
. withi
p. = i - Z k.,1 > o,i-l
IM ^ #p, 0" = 2, 3, •••).
/Aen |x4| g JT(f - 1, 2, •••).
Throughout the paper, we use the notation
In addition, the subscripted capitals Cu C2, • • • will always denote positive real
constants.
4. The Trapezoidal Product Integration Method. We replace s)f(s) in
Jo (r,. — i)
by the piecewise linear function
Mi,(tits) = + 1 - sykt.J, + (s - ^)£,,)+17, + I]/Af
[t, ^ s rS ti+1 0 = 0, l)},
and thus obtain the following lower triangular system for /< (/ = l, ■ • • , /):
(4.1) £ = gi/(A(1->(«)) - /0A:,.ofP.- (/ = 1, , /),
where
*(*) = —1-t-^— ,I — a 2 — a
<*(<*) A<2_° Jo fa0(a)
.(I+DAl
^' rfl + /0(a)
= (/ + l)2"* - 2l2~a + (I - 1)2~° (#—-.»,«•• . 7 — 1).
<6(a) A/ VJc,-nil ,« t" I
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
180 richard weiss
0(a) At ° f
/ _L_ (/-« _ (/ _ _ (L-Jl (/- _ (/ _ !)"«)}1,2 — a 1 — a J
= T71-;0(a) 1,2 — a
(I - 1. ••• . I).
For the starting value /0, we take
f0 = U - Hm (gW^^Xl ~ «)Ao.o,I-.0
which exists, whenever (1.1) has a continuous solution. It follows from (1.2) that
(4.1) is nonsingular.
The numerical process (4.1) defines the product integration analogue of the
trapezoidal rule applied to nonsingular first kind Volterra equations {see Linz
[10]}. We shall refer to it as the trapezoidal product integration method.
As in [7], we first investigate the convergence of (4.1) for the case k(t, s) = I.
Theorem 4.1. 7/7(0 's three times differentiable and f"(t) is boundedfor 0 g t g T,
then, for a £ [au 1) with ax = 0.2117, (4.1) is convergent of order 2.
Let /4 ,(s) denote the piecewise linear function interpolating f(s) at the grid points,
fitO) - [(',+> - tilt + (/- tdh.iVAt r, + I 0' = 0, •••,/- 1)},
and let
eA((i) = /a<(s) - /(*).
By subtracting
Ar1_o0(a) \Jo (/, - Jo (/, - a)« /
from (4.1), we obtain
(4.2) = --r^— f & " ^< (/ = 1, ••• , /)•At 0(a) Jo (,,. - S)
Finally, subtraction of (4.2) from (4.2), with /' replaced by i -f- 1, yields the basic
error equation
(4.3) e«+, - £ 6,fl,_, + fc, (j = 1, • • • , 7 - 1),j-i
with
«0 = Wo - Wi = 3 - 22-°, a, - IF, - r,+, (I - 1, •••,/- 1),
0(a) Af1 "
(4.4) ft = *-g p (—L_ - —-5"-^*« *.(/, + 1 — s) i-o J<, — s) (f<+, — s) /
ft2 = £o(fF, - Wi+1).
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
the generalized abel integral equation 181
Before we give a proof of the theorem, it is first necessary to obtain some preliminary
results.
Lemma 4.1.
\ß)\ is cAf—r- (/= l, •••,/ - i),
(4.5)
\ß2\ g cMrl~a 0 = l, •••,/- i), if |e0| ^ c3 a*2.
Proof. By a standard theorem on Lagrange interpolation
= Us - ',)(',+. - *)/"(/,) + r,(s)
[t, g s g rI+1 V = 0, •••,/- 1)},
with
(4.6) M»)| g 17*3Ar3, F3 = max /"'(/).ieio. t]
By substituting (4.5) into (4.4), we obtain
ß) - Ar3-°(/"(r,)7 - z HWfw) + *<•
where
1
2 At Jo
0, = -~- f s'At - s)(-) ds,2 At3-" Jo \l At + s)a ((/ + 1) Ar + s)a>
jjBr_i^_^irp-l_aJ« (r.+1 - *)- f=i J,t Xu - s)° (ti+1 - s)aJ
It can easily be verified that, by (4.6),
(4.7) \6i\ g C4 At*~a
and that
(4.8) 0 < 0o, 0 < 6, < CJ
(f - 1, 1),
».tI c.r'.
/ ^ i.
f ^ l.
(s) cfa.
(4.9)
By partial summation,
z /"(*,)««_*-, - f(f«-i) z »i + z (/"('.-,-,) - /"('<-»)) X z «ij-0 1-0 »-l !-»
(/ = 2, ■•• , / - 1)
and, using (4.8),
(4.10) z (T(t<-,-i) - /"('<-,)) X z »i rS C7 Ai/1"" (i = 2, • • • ,7-1).
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
182 RICHARD WEISS
Combining (4.9) and (4.10), we obtain
/"(/<>Y - Z /"(*,)*«-,-.
(4.11) ^ f"(tJy - Z 0«)\ 1-0 /
g F2C6ra + 7F3 A/ + C7 Afi
+ l/'U) - /'U-i)| X Z -5, + C7 A//1"
(/ = 1, ••• , / - 1), Fa = max /"(*)•• eio.ri
Since
Ar = T/I rg T/i (i = 1, ■ •• , /),
it follows from (4.7) and (4.11) that
\ß]\ £ c, A/3-ar" («• = i,... , / _ i).
The second inequality of Lemma 4.1 is an immediate consequence of the asymp-
totic expansion of (W, — W,+1). □
Proof of Theorem 4.1. Using Lemma 4.1 it is clear that
(4.12) ^ CsAfra (I - 1, •••,/- 1).
Returning to (4.3), we note that
(4.13)
(4.14)
(4.15)
Z*i « 1 - Wt\ 1.
a. > 0, / £ 1,
0 g a0 < 1. a G [«o, 1),• • ■
-1 < a0 < 0, a G (0, a0), a0 = 2 - == 0.4150.In 2
Substituting for e, in (4.3), we obtain
(4.16) e<+1 = Z + 6. + «o*<-i (i = 2, •••,/- 1),i-i
with
a0 = 0, a} = a, + a0a'-i, /SI,
and
al > 0, a G (0, 1),
a) ^ 0, Ü 2, a G t«i, 1), «i =4= 0.2117,
al < 0, a G (0, «,).
The latter is a consequence of the structure of {a,(a), / ^ Oj. From (4.13),
(4.17)
Z a\ = 1 - (W, + ao^-i), » ̂ 2,
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
the generalized abel integral equation 183
and, since
eg
W, = a, + Wt+l = £ a„ i ^ 0,
it follows from (4.17) that
W< + flo^.-i = Z) (a. + «ofl,-i) = Z «i > 0, i ^ 2, a G 1).
Consequently,
(4.18) £a} g 11-0
since behaves asymptotically like
It follows from (4.2) that
(4.19) k| g C10Af2, Id -g C10A/2.
We are now in a position to apply Lemma 3.1 to (4.16). On the basis of (4.19) together
with (4.17), (4.18) and (4.12), the conditions of the lemma are satisfied and the requiredresult
W ^ C„A? (/ = 0, ••• , I), a G [«„ 1)
follows. □
Note 4.1. Using Lemma 3.1 and (4.13)—(4.15), we can prove second order con-
vergence for (4.1) with a G [«o, 1) without making a substitution of the type (see
(4.16)} introduced in the proof of Theorem 4.1. On the other hand, if we substitute
for e,-_2 instead of e, in (4.16) using (4.3), we have
i
(4.20) ei + 1 = £ e,ö2_, + bt + «o0,-i + «5*,-, (i = 4, •••,/- 1)i-i
with
2 A 2 1 2n 2 111 ■ *w 1aa = 0, «! = au a2 = 0, a( = at + a2"(-3. ' s 3.
It can be shown numerically that (4.20) satisfies the conditions of Lemma 3.1 for
a G [a2, 1) with a3 == 0.1292. It follows that, if the proof of order 2 convergence
for all a, 0 < a < 1, by means of Lemma 3.1 is possible, it will involve a complex
inductive argument. However, because our proof already covers the case a «■ J,
which is the most important one in application, an extension of the result of Theorem
4.1 will not be pursued further in this paper. Numerical experimentation indicates
that second order convergence holds for all a G (0, 1). This is illustrated in Section 6.
Since the extension of Theorem 4.1 to a general k(t, s) can be established using a
technique analogous to that of [7] for midpoint product integration, we only state
the result.
Theorem 4.2. If(i) /(/) satisfies the conditions of Theorem 4.1,
(ii) k(t, s) is twice continuously dijferentiable with respect to t and s for
0 ?S s -g t ^ T,and
- Col" (i = 2, 1), a G [a,, 1),
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
184 RICHARD WEISS
(iii) d3k(t, s)/ds3 and d3k(t, s)/dt ds2 are bounded for 0 rg s rg t rg T, then, for
a G Wi, 1), (4.1) is convergent of order 2.
5. Computational Errors. During the actual application of algorithm (4.1),
the values {/„, • ■ • , //} instead of {7o, • • • , //}, which satisfy /„ = f0 + t;0,
£ Jfc.i Wt-t = gi/(Atl~a<p(a)) - }jc(,*Wi + Vi a = 1, • • •, D,1-1
are determined, where the perturbations {Vi, i = 0, • • • , I) represent the effect of
rounding and truncation errors. It follows from Lemma 3.1 that
(5.1) I/,- - 7.| ^ + h d = 0, • • • , 7), a G \fii,i 1),
where = const > 0 and |%| £ 9 (/ ■» 0, • •J), However, since the application
of Lemma 3.1 cannot yield a sharp estimate in this case, (5.1) represents a conserva-
tive bound. Numerical experimentation indicates that
max \ f( — f{\ g K2ri,os.sr
where K% is a positive constant depending on T. This is the usual behaviour associated
with computational errors in the numerical solution of Volterra integral equations
of the second kind; equations of the first kind tend to be less well conditioned.
6. A Numerical Example. The trapezoidal product integration method was
applied to the following equation:
n ' v,. IxFiiU 2 - a; it) - ^,(1; 2 - <x; -it)] = f /C?) - ds,(6.1) v '
0 £ * £ «r,
where iFj is Kummer's hypergeometric function {see Erd61yi [11, Section 13, p.
189]}. (/ is the imaginary unit.)
The solution is f(t) = sin t.
The errors obtained for a = 0.5, for which convergence is guaranteed by Theorem
4.1, and for a = 0.05 are listed in Table 6.1. In both cases, the error satisfies the
predicted At2 dependence.
Remark. All numerical results presented in this paper were computed in double
precision arithmetic on the IBM 360/50 computer at the Australian National Uni-
versity.
7. A Third Order Method. The existence of higher order methods for (1.1) is
not immediately obvious. So far, convergent methods of order greater than 2 have
not been found for the first kind Volterra equation
(7.1) g(0 = f k(t,s)Ks)ds,Jo
where k(t, s) satisfies (1.2) {see [10]}. Equation (7.1) can be considered as a special
case of (1.1) with a = 0. Linz [5] has shown numerically that fourth order methods
can be constructed for the equation
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
THE GENERALIZED ABEL INTEGRAL EQUATION
ino
•o
I
8
00
\-p
<
in
o
II
Ö
ioi
w
o
inI
w
COo
CT.
IW
CMO
IW
inIw
oo
inlw
cooID
Iw
in
•a1Iw
co
on
CO00
iniH
inin
l
CMro
lH
COCN
inIw
ino00
I
00o
IW
enCMin
lw
oninino
COCOcnin
inI
W
"=rinmo
inIa
cnCM
inI
W
inino
inI
W
t-~cnCnin
inI
w
CT>CO00
"31
Iw
ID
COo
coCMCM
IwCA
mCO
CM
ID
CMCMCM
IW
CMO«*o
inIW
inin
Iro1w
COCM
OlCO
IW
oin
I
onCO
I Iwr-r-CMo
IH
VDCM
COI
w
o
roI
W
ID
cn
roI
W
roI
H
COIDrHCM
Iw
in
W
ini
w
inI
W
mIw
oo
inI
H
in
CM
cn
CM
inla
inoncnCO
lH
ID
Iw
TTon"*CM
Iw
toooCM
COo
IW
i—ICOr-co
Iw
r-roCOID
IW
CM
on
I I
oo
Iw
r-inoin
CO
lw
COI
H
cnoo
Iw
Iw
ooninCM
ID
w
in
cn
cn
inI
H
ID
r~cn
IW
ro
IW
inI
W
in-a1o
inI
a
r-CM
I -W
inCMIDf-
IW
oonCOCO
00
HID
CO
r-
cno
oin
IDIa
inCMCOCO
o•a-cn
IDI
H
CO
IDCO
iniw
■a-in
IW
EhEhEhEhEhEhEhEhBEh<<<<<<3<1<I<<
cMco^iniDr-coono
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
186 RICHARD WEISS
HS
Eh<
+)
o
II
a
r-i
IOIw
<nco
in
W
^pI
W
r-ino
coI
W
r-
lm
m
<J5
IW
ini
W
Eh<
iniW
o
IH
incm
CO
i
W
mcoin
w
corocor-
10
lw
inIw
uoo(o
EH<Ol
IH
r-
iW
ocm
I
r-r-
r-i
w
oIN
IKl
r-mr-
ml
H
coIW
iW
or-
cmi
i
lw
Iw
cmco
W Eil
IS
Iw
Iiniw
ococn
inl
W
cmI
Iw
incoin
l
ini
w
coI»
coI
IW
cm^p
cocm
IW
cnco
iw
cm
coI
w
ino
co
coIw
cmI
inl
r-l
H
in10o
r» 00i l
W W
cm r»rH cmr-i cmOl cm
r-I
H
oco
r-l
M
10«*
in<oTT
cmI I
10I
10Iw
coin
iw03co
H W
co >*on cmin oin co
Eh<3co
Eh<
iW
cm
irjI
W
CO
Eh
in
ioi
w
IOr~co«*
idIw
o00
EH<IO
cm
coI
inIw
COCnCM
EH<a
mla
coidcoin
col
EH<CO
IW
^poo
r-I
cmCnOlin
ini
iniw
iw
CO
r-
cnI
orocm
IO IOI I
W H
in
Eh<CTi
IW
l
IW
rocn
inl
ini
W
i
00 00 f»I I I
H H W
or-.roon co corH t- cmrH «3 O
r- r- r-l l i
W W H
in
IOI
w
cm I1 rHvo mio «tf <*co on ^
io
inl
W
coo
Eh<orH
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
THE GENERALIZED ABEL INTEGRAL EQUATION 187
T3
sa
■h+>c,O Oo
II
a
o
idIw
in
W
oinCO
IW
IMin
Eh<
ißI
H
cmoocm
inlw
r-l»ooo
IW
idididco
W
id
On
EH<cm
IW
oidid
IX)
Iw
r*i1
cocn
IDO-l-a-cn
idI
W
cocmcmcm
id
w
lcmI
coI
inlH
iniw
ml
W
o ido r-ir- ino o
lW
op«co
inlW
cm cnI
cmI
W
cm
Iw W W
rH oo cmT rHr- cm
cmcm
inI
W
coon
ro cnI I Iwww
co r-
co oidco
IH
1"or-
IH
Oin
co
cm
En Eh< <co •o'
Eh<in
Eh Eh<1 <id r-
Eh<00
I
oco
coI
IX»
EH<on
id
H
I
inl
W
o ino in
o in
co
^ ^1 ^J"
I I IWWW
o i-l r-co on oooo r- oo co
cmI
coI I I
WWW
in on incm in onco on oncm oo o
<1o
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
188 RICHARD WEISS
g(t) = f 2 m,1/a ds, 0 rg t g r <Jo (' — i )
However, it is not possible to infer the existence of higher order methods for (1.1) on
the basis of this existence, since the integral operator
AM = f , 1(S\adS> a<l>Jo (t — S)
is a compact operator from C[0, T] to C[0, 7], while the operator
is not.
We verify numerically the existence of higher order methods for (1.1). In fact,
we construct a third order one. For simplicity, we take k(t, s) = 1. The generalization
is straightforward. We assume I = TAt is even and, generalizing the technique used
in the development of the trapezoidal method, we approximate j(s) by a function
MÄ t(s) which is a quadratic polynomial in each of the intervals [t2j, t2i+2] (J = 0, • • • ,
1/2 - 1):
A/a,(s) = f2i +s — hi
(J21 + 1 /21) ~T"Q — t2i)(s — f2f+i)
(?2i + 2 2f2j + l + /2;,
rg s rg f2i+2 0" = o, ••• , 1/2 - i)j.
By substituting M4,(s) for /(s) in (1.1), we obtain
Jo ('21 + 1
i + 2 — /Jo
(7.2)MAt(s)
ds
ds
(j = 0, • • • , 7/2 - 1)
('21 + 2 s)
which, for each j, is a nonsingular two by two system for J2i+l and /2l+2. Thus, (7.2)
is a block by block method in the sense of [10]. Evaluation of the integrals in (7.2) and
division by At1'" yields
£_,/2 + F.Ji + G-J0 = gjAt1-",
£0/2 + Fjx + Go/o = g2/Atl~",
E-if2i+2 + F.iJu+i + Z Ji Fu+I_, + f0G2i-i = g2i + 1/At1'
0 = 1. 1/2 - 1),
E0f2i+2 + Foht+i + Z h Wu+x-t + foG2i = g2i+2/A/1 '>'-i
70 = lim GKOf-'Xl - a),1-.0
F2j = F2,_i (/ ^ 1); F2i + i — G21-1 + £21 + 1 (/ ^ 0);
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
THE GENERALIZED ABEL INTEGRAL EQUATION 189
W2l = F2l (/ ^ 1); W2M = G2l + E2l+2 (/ ^ 0);
1 2 \3 - a T2 - a/'G
= LX<J, a)(l + fy + i/) - L2a, a)(§ + j) + \UO, a) 0 S 0)
F, = -IM a)(2j + /) + L2(J, a)(2 + 2/J - L3(j, a) (j £ 0)
G,- = 1,0, a)G7 + If) ~ L2(J, a)(| + j) + \L3(J, a) <J ̂ 0)
1*0, a) = (0' + 2)'-" - - a);
L2U, a) = (0' + 2)2-" - /""°)/(2 - a);
£30, = (0 + 2)3~a - /-)/(3 - a).
This method was applied to (6.1). The errors for the cases a = 0.9, a = 0.5 and
a = 0.1 are tabulated in Table 7.1. Third order convergence holds for a = 0.9 and
a = 0.5. For a = 0.1, a strict At3 dependence is no longer apparent. However, the
process is still convergent—no explosion in the errors is observed. The numerical
approximations oscillate about the true solution, and the oscillations increase as
a tends to zero. The simplest interpretation that can apply to such behaviour is
one of increasing potential instability as a decreases.
A rigorous analysis of the convergence properties of this method cannot be given
here. The problem is the lack of accurate estimates for the solution of the lower
triangular system of equations which determines the behaviour of the error.
Acknowledgment. The author wishes to thank Dr. R. S. Anderssen for his as-
sistance and helpful comments during the preparation of this paper.
The Australian National University
Computer Centre
Canberra, A.C.T. 2601, Australia
1. Gerald N. Minerbo & Maurice E. Levy, "Inversion of Abel's integral equation bymeans of orthogonal polynomials," S1AM J. Numer. Anal., v. 6, 1969, pp. 598-616. MR 41#6426.
2. R. M. Fortet, "Les fonctions aleatoires du type de Markoff associees ä certaines6quations lineaires aux derivees partielles du type parabolique," /. Math. Pures Appl., v. 22,1943, pp. 177-243. MR 7, 19.
3. J. Durbin, "Boundary-crossing probabilities for the Brownian motion and Poissonprocesses and techniques for computing the power of the Kolmogorov-Smirnov test," /. Appl.Probability (To appear.)
4. A. Young, "The application of approximate product-integration to the numericalsolution of integral equations," Proc. Roy. Soc. London Ser. A, v. 224, 1954, pp. 561-573.MR 16, 179.
5. P. Linz, Applications of Abel Transforms to the Numerical Solution of Problems inElectrostatics and Elasticity, MRC Technical Summary Report #826, University of Wisconsin,Madison, Wis., 1967.
6. B. Noble, Lecture Notes, 1970.7. R. Weiss & R. S. Anderssen, "A product integration method for a class of singular
first kind Volterra equations," Numer. Math. (To appear.)8. G. Kowalewski, Integralgleichungen, de Gruyter, Berlin und Leipzig, 1930.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
190 RICHARD WEISS
9. L. V. Kantorovic & V. I. Krylov, Approximate Methods of Higher Analysis, 3rded., GITTL, Moscow, 1950; English transl., Interscience, New York; Noordhoff, Groningen,1958. MR 13, 77; MR 21 #5268.
10. P. Linz, The numerical solution of Volterra integral equations by finite differencemethods, MRC Technical Summary Report #825, University of Wisconsin, Madison, Wis.,1967.
11. A. Erdelyi, et al., Tables of Integral Transforms. Vol. II, McGraw-Hill, NewYork, 1954. MR 16, 468.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use