a sine approximation for the indefinite integral€¦ · received april 23, 1981; revised march 10,...
TRANSCRIPT
MATHEMATICS OF COMPUTATIONVOLUME 41. NUMBER 164OCTOBER 1983. PAGES 559-572
A Sine Approximation for the
Indefinite Integral
By Ralph Baker Kearfott
Abstract. A method for computing fa f(t) dl, x = (0, I) is outlined, where f(t) may have
singularities at t = 0 and í = 1. The method depends on the approximation properties of
Whittaker cardinal, or sine function expansions; the general technique may be used for
semi-infinite or infinite intervals in addition to (0,1). Tables of numerical results are given.
1. Introduction and Summary. Here we present a basic method and experimental
results for approximating Fix) = /0V fit)dt, x E (0,1), where/is smooth, but may
have singularities at t = 0 or t = 1. The method is based on approximation proper-
ties of sine functions
iir/h)ix — kh)
With minor modifications, the method may also be employed to compute jx fit)dt,
t E ia,b), where a, b or both a and b are possibly infinite.
Our underlying formula is similar to formula (3.36) in [4], given there without
proof. Our formula is considered in detail here, is proven, and is backed by
numerical experiments. (See Section 4 for more discussion.)
In Section 2 we give the relevant properties of sine functions, assumptions, and
basic techniques. (Such techniques appear in more detail in [4] and elsewhere.) In
Section 3 we present and verify our integral approximation formulas. Four tables of
experimental results and conclusions appear in Section 4.
2. Underlying Properties and Assumptions. A general review of sine functions and
their uses has recently been given by Stenger in [4]. We therefore only outline
properties important to our present goals, and refer the reader to [4] for further
references.
Approximations on (0,1 ) are obtained from corresponding approximations on
R = (-co, oo) via a conformai map. To be approximable on R, /£ C°°(R) must
obey certain analyticity and boundedness conditions in a strip in the complex plane
C which contains R. Through the conformai map, we obtain a corresponding
"eye-shaped" region (cf. Figure 4.1, p. 185, of [4]) containing the interval (0,1), in
which our integrands must obey certain analyticity and boundedness conditions.
With this in mind, we first list properties over R.
Received April 23, 1981; revised March 10, 1982.
1980 Mathematics Subject Classification. Primary 65D30; Secondary 41-04, 41A25, 41A55.Key words and phrases. Indefinite integrals, singular integrals, quadrature, sine functions, Whittaker's
cardinal function, approximation theory.
^1983 American Mathematical Society
0025-5718/83 $1.00 + $.25 per page
559
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
560 RALPH BAKER KEARFOTT
Defintion 2.1. Let d > 0 and let 6Üd denote the open strip
(2.1) %={zEC:\lm(z)\<d).
Let B iaf}d) denote the family of all functions/ that are analytic in s\)d and which
satisfy
(2.2) ¡d\fix + iy)[dy-+0 asx-±oo
and such that
Wp
< oo.(2.3) Np(f,%) = ma_Uf\fix + iy)\Ddx)j + (/j/(x - iy)f dx)
We denote B^d) by Bi^d); in all results below, p = 1 for simplicity unless
otherwise stated, although most of these results generalize to arbitrary p; cf. [4,
Section 3.1].
If/ E Bi^j), we approximate/by a truncated " Whittaker cardinal series"
(2.4) f(x) = CNif,h)(x)= 2 f(kh)S(k,h)(x),k = -N
where Sik, h)ix) is as in (1.1). We state the following without proof:
Theorem 2.1. (See [2, pp. 237-238], and [4, pp. 177-178].) Suppose f E Bpi%)for
some p E [1, oo), and suppose
(2.5) \f(x)\<Ce-* forallxER(t.e.,\f(x)\= 0(e-"M))
for some C and a > 0 dependent on f. Choose h = [7rd/(aN)]]/2. Then there is a C"
dependent on f, but not on N, such that
(2.6) H/U) - CN(f, h)ix)\\x < c'A"/2e-(WaA,»'/2,
where || • H^ is the supremum norm over R.
Here we need not only to approximate functions /, but also their integrals. We
have
Theorem 2.2. (Trapezoidal Rule; cf. e.g. [2, p. 229], or [4, p. 178].) Suppose
f E Bi^Ùj), suppose f satisfies (2.5), and choose h' = [2-nd/aN]l/2. Then
(2.7)r ¿,-(2T7ctaN),/2c,efXf(t)dt-h' 2 f(kh')
-oo k = -N
for some integral approximation constant C, dependent on f but not on N.
To obtain approximations over (0, 1), we make use of the following conformai
map:
(2-8) v(,) = kjg(T4_), 9'(z)=^~y
For any d such that 0 < d < ir/2, q> maps the region
(2.9) %={z:\arg[z/(<-z)]\<d}
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
A SINC APPROXIMATION FOR THE INDEFINITE INTEGRAL 561
onto ^j where 6ffd is as in (2.1). (Note that 9^ consists of two circular arcs
intersecting symmetrically with an angle of 2d at 0 and 1); (cf. [4, p. 185].) Define
(2.10) z = xpiw) = no-\w) = j^, xP'iw)=xPiw)[l-xPiw)].
Suppose/is defined and analytic in the interior of ®ùd, and define
(2.11) f(^)=f(^))xp(w)[l-xpiw)]
for w E %. Then/ G Bi^d) provided
(2.12) lim inf f \f(z) dz\< oo,
where Q is some curve; we obtain (2.12) from (2.2) and (2.3) by applying the change
of variables z = xpiw) to the integrals in (2.2) and (2.3).
Suppose, for example,/has singularities at z = 0 and at z = 1, but/is analytic in
°0d, and / is continuous on 3'?)</\{0,1). Then, by examining the integrals over
portions of the curves S near z = 0 and z = 1 and the integrals over portions of G
away from z = 0 and z = 1 separately, one can show (2.12) holds if and only if
f \f(z)dz[< oo.
Suppose further that/is such that we may take d = it/2, so 9^ = [z:\z — {\= {).
Then xp maps the line (z : Im(z) = tt/2} to the semicircle (|z — {-1= {, Im(z) > 0}
and the line {z : Im(z) = -m/2) to the semicircle {\z — \\= \, Im(z) < 0). Condi-
tion (2.12) then becomes
(2.12)' Se\f(z)dz\=\j2;\f(\ + \e«)\dt
= C"/4 \f(eiucos(u)) | du + r/A |/(1 - eivcosiv)) \dv<*o.Jir/4 J-tt/4
Condition (2.12)' then holds for those/which satisfy
(2.13) f(z) = O(|z|°o) asz-^0, and f(z) = 0(\1 - zf<) asz-*l,
where a0 > -1 and a, > -1, and which are continuous elsewhere on the closure of
tf)d and which are analytic in tf)d.
Corresponding to Theorem 2.1, we have
Theorem 2.3. Suppose f is analytic in some tf)d as in (2.9), suppose f satisfies (2.12)
ior (2.12)'), suppose
(2.5)' |/(jc)|<Cx"(1 -x)p, 0<x<l,
for some ß > -1 and some C dependent on f and set h = [Trd/aN]x/2, where
a = ß + 1. Then
(2.6)' /(*)" 1 f(Hkh))xP(kh)[l-xP(kh)]S(k,h)(<p(x))k = -N
< C'Nx/2e'(,rdaN)'/2
for some C dependent on f but not on N.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
562 RALPH BAKER KEARFOTT
Similarly, we have
Theorem 2.4. Suppose f is analytic in tf)dfor some d, suppose f satisfies (2.12) (or
(2.12)'), suppose f satisfies (2.5)' for some ß > -1 and some C dependent on f, and set
h' = [2ird/aN]l/2. Then
(2.7)' Cf(t)dt-h' 2 f(Hkh'))xP(kh')[l - xp(kh')]•'n ,k = -N
for some C, dependent on f but not on N.
^C,e -(27TÍ/OV)''
3. The Indefinite Integral Approximation Formulas. Analogously to Section 2, we
will first obtain an approximation procedure for integrals of the form
(3.1) F(w) = f f(t)dt,
where / E Bi fyd ) (cf. Definition 2.1). We will then use the conformai mapping
procedure to obtain approximation rules for integrals of the form
(3.1)' F(x) = f X)fi<pit))dt,
where <p and xp can be as in formulas (2.8) and (2.10), respectively. We begin with the
following lemma, which is a direct consequence of Theorem 2.1.
Lemma 3.1. Define Fiw) as in (3.1), suppose F E Bpi0l)d), for some p E [1, oo),
suppose |F(w)|s£ Ce""w for some C and a > 0 and all w E R, and set h =
[■ïïd/iaN)]]/2. Then
(3.2) /(Odtí N
1(k = -N
f/0>di Sik,h)iw) < C'A/l/V'"/o,/V)'
for all w E R and for some indefinite integral approximation constant C¡ which does
not depend on N.
Lemma 3.1 does not apply to general integrals F as in (3.1), since in general
fHc00fit)dt ¥= 0, and formula (2.5) is not valid. For such F, we consider the related
function
(3.3) G(w) = f_ M*- f f(t)dt
-c f(t)
?-—■'' -H e
ae~'Uf) dt i\(t)dt.
where
(3.4) *(')=/(')ae
(e"'+l)
(ea'+ iy
Uf), Uf) = ff(t)dt.
We then have
Lemma 3.2. Suppose |/(w)|< Ce "M for some C and a > 0 and all w E R, and
suppose G is as in (3.3). Then there is a CG such that | C7(w) |< Cce"aW for all w G R.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
A SINC APPROXIMATION FOR THE INDEFINITE INTEGRAL 563
Proof. We have
|G(W)|.
,aw/2
,aw/2 _|_ £-aw/2f f(t)dt~Ix(f)
+-aw/2
/w f(t)dteaw/2 + e-
We will consider each factor in the two terms on the right as w -> -co and as
W -» +00.
Case 1, w -» -co. The second factor in the first term is bounded and approaches
-Ixif), while the first factor in the first term is bounded by e"a|M'1. The first factor in
the second term is bounded by 1, while the second factor in the second term is
bounded by Ce-aW/a.
Case 2,w-> oo. The first factor in the first term is bounded by 1, while the second
factor in the first term is equal to -J™fit)dt and hence is bounded by Ce'^/a.
The first factor in the second term is bounded by e"aW, while the second factor in
the second term is bounded and approaches Ixi f ).
Combining the above, we may take
Cc = C/a + max sup / /(/) dt - Ijf) , sup / fit) dt\ w«0 l-oo w>0 l-oo
~C/a + \IJf)\.
Combining Lemma 3.1 and Lemma 3.2 immediately gives
Theorem 3.1. Suppose /G Bi%), suppose |/(w)|< Ce"aW for some C and all
w E R, and suppose G E Bpi^d) for some p, where G is as in (3.3). Define
(3-5) FN(w) fit)ae
,aw/2
+
(e°" + 1) J
/CO
f(t)dt,
/OOf(u) du dt\S(k,h)(w)
e-aw/2 _|_ e«w/2
where h = [<nd/iaN)]x/2. Then there is a C'„ independent of N such that
/w fit)dt-FNiw) (J']yW2e-(irdaN)
for all w E R.
In practice the quantities
rkh
f /(<)-ae
ie°" + 1) -«C f(u)du dt= fkhg(t)dt = G(kh),
and the quantity /^/(u) du = Ixif) must be approximated. Below, we show how
to do this so that the order of approximation in (3.5) remains OiN2e'^daN) ).
First observe that linear combinations of functions in Bitf)d) are also in Bi^d);
also linear combinations of functions satisfying (2.5) satisfy (2.5) (with a different
C). Thus, we deduce
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
564 RALPH BAKER KEARFOTT
Lemma 3.3. Suppose f E Bi^), suppose |/(w) |< Ce~aM for some C and a > 0 and
all w E R, and suppose g is as in (3.4). Then g E Bi^), and there is a Cg such that
\giw) |< Ce"* for all w E R. Thus, g and jfx git) dt may be approximated as in
(2.6) and (2.7), respectively.
Let CNig, h)it) = 2j=.NgiJh)Sij, h)it) be the approximation to g. We use
CNig, h)it) to approximate the quantities Gikh) in (3.5).
Lemma 3.4. Let f G Bi%), suppose ¡fit) \< Ce"""1, t G R, and set h =
[■7rd/iaN)]l/2. Suppose g and G are as in (3.3) andi3.4), and define
N
(3.6) GNiw)= f CNig,h)it)dt= 2 g(Jh)f S(j,h)(t)dt,J-Ml, ., J-Wl,-Nh j = -N -Nh
where -Nh < w < Nh. Then there is a constant C dependent on f such that | GNiw) —
G(w)|< CNe-l"daN)>/2 for all w E R.
Proof.
\GNiw) - G(w)\*z\G(-Nh)\+ f \g(t)-CNig,h)(t)\dt.J-Nh
But | G(-/YA)|< Cce'aNh = CGe~(,näaN)i/\ from Lemma 3.2. Also
f [git) - CN(g, h)(t)\dt < (NH + w)C'Nx/2e-("dttN)W1 < 2N^2hC'e^daN)W1J-Nh
= CNe-^daN)>/\
from Lemma 3.3. Thus
|Gw(w) - G(w)\ < CGe-("daN>'/2 + CNe-("daN)l/1 < CNe^daN)W1
for some appropriate C.
We now replace Gikh) by Gikh) in Theorem 3.1.
Lemma 3.5. Suppose the assumptions of Theorem 3.1 hold, but
rkh
}(kh) = f fit)ae"
/OOJW) du dt(ea' + 1) J~<*
in (3.5) is replaced by GNikh) as in (3.6), to obtain
(3.5)' FNiw)= 2 GNikh)Sik,h)(w) + - J™'2 aw/2C fit)k = -N e
Then there is a CF independent of N such that
dt.
\f fit) dt - FN(w) CFN2e-(m/aN)'/2.
Proof.
/W A I fW Afit) dt - FNiw)U\j fit)dt-FNiw) +\FNiw)-FNiw)\,
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
A SINC APPROXIMATION FOR THE INDEFINITE INTEGRAL 565
where the first term on the right is bounded by C',Nx/2e (7"JaN>'/\ by Theorem 3.1.
On the other hand,
\F„(w)-FN(w)\ 2 [G(kh)-GNikh)]sik,h)iw)k = -N
N
< 2 \Gikh) - GNikh)\^2NiCNe-^daN)>/1)k = -N
< C N2e-<«daN)'/2
for some constant CF dependent on/but not on N.
For the main result, we must replace Ixif) = ffx fiu)du in (3.5)' without
changing the order of approximation.
Lemma 3.6. Suppose the assumptions of Theorem 3.1 hold, but in (3.5)' we replace
U/) = /-»/(')* by INif) = hl^^.fikh), where h = [t7¿/(«A')]i/2. Then theconclusion of Lemma 3.5 is still true.
Proof. If there are a C and an a > 0 such that |/(/)|< C<?~a|1, / G R, then, if
a' = a/2, |/(r)|< Ce-"'n,t E R. Thus from Theorem 2.2,
(3.7)
Define
(3.8)
and define
and
lU/)-'v(/)l<c/e-<^">'/\
lv(0= 2i = -N
f(jh)ae
a ¡h
Mf)(eajh + l)2
Gfl(») = f C^g^AKí)^
S(j,h)(t),
-Ml
,aw/2
(3.9) FNiw)= 2 GN(kh)S(k,h)(w)+ -' ** ' /v/ _-aH>/2 i .aw/2
a- = -a/ e 7 -t- e 7/*(/)•
Then
(3.10) \f f\t) dt - FN(w)\ *¿f f(t) - FN(w)+ \FN(w)-FN(w)\,
where FN is as in (3.5)'. By Lemma 3.5, the first term on the right of (3.10) is
bounded by CFN2e-^daN)i/\ while
(3.11) \FN(w)-FN(w)\k = -N
+
2 [GN(kh)-GN(kh)]s(k,h)(w)
Uf)-INif)\,,aw/2
e-aw/2 _j_ £aw/2
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
566 RALPH BAKER KEARFOTT
where GN is as in (3.6). The first term on the rieht of (3.11) equals
(3.12)
aea jhN N
2 2k=.N [j=.N (e«i* + l)2
[/*(/) - Uf)]fk"s(j, h)(t)dt\s(k, h)(w)
while the inner sum in (3.12) equals
aeajh k-j SÍn(7T?)(3-13) Ak,N = h 2 -^-—2\iAf)-Uf)\r^P-dt.
/ = _.v ie'fh + j) N-j
But AkN is a trapezoidal rule approximation, as in Theorem 2.2, to J^x Xk Nit)dt,
where
(3.14) \k,Aw)=[lN(f)-Uf)]ae
(eaw + l)2 Ik-w/h SÍn(77-í)
N-w/h 77/dt
Sk-w/h sin(w/)
N-w/h <!ttdt sech
[(!)-]■
where h' = i<nd/aN)x/1 = i2-nd/a'N)x/1 for a' = a/2. But the magnitude of the
integral in (3.13) is bounded by
2 r"sin(i)
thus
2 fsm(t) .-sup / ——dt<2;
(3.15) |A*,„(hO|< 2a|/y(/) 7°°(/)l <2a\INif)-IJf)\e-W.i£aw/2 _|_ e-aw/2y
It is also possible to verify that Xk Niw) E Bityj). Thus, as in Theorem 2.2,
(3.16) |/°V*(odt- Ak.N C-eA 0-(TidaN)wl
dt
for some constant C,A dependent on /but not on N. But, by (3.7)
(3.17) \fykMdt\<±Mf)-iM)\[fy&[(%y
= \IN(f)-Uf)\<C,e^d°^/2.
Combining (3.17) and (3.16), we see that the quantity in (3.12) is bounded by
N
(3.18) 2 (C, + cf)e-<"daN)'/2 = 2(C, + C,A)Ne^daN)'/2.
k = -N
Combining the bounds (3.18), (3.7), and (3.11) now gives
(3.19) \FN(w) - FN(w)[< 2(C, + CfA)Ne-("daN)'/2 + C,e^daN)W2.
We obtain the conclusion in Lemma 3.6 from Lemma 3.5, 3.10, and 3.19.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
A SINC APPROXIMATION FOR THE INDEFINITE INTEGRAL 567
We summarize the above in the following
Theorem 3.2 (Main Result). Suppose f G Bi%), and |/(w) |« Ce"aM for some
a > 0 and C, and all w ER. Suppose further G E Bi^j), where G is defined in terms
°ff by (3.3). Define aq, q an integer, by
(3.20) aq-^——dt.
Set h = [trd/iaN)]]/2, and approximate f_wxfit) dt by
(3.21)
FA») f(jh)~ae a i h
(eajh + I)2Mf) (ar,+J + ak_J)\S(k,h)(w)
,aw/2
e-aw/2 + £aw/2 nU ),
wheretNif) = h%%=_Nfikh). Then there is a constant CF independent of N such that
\f fit) dt - F„iw) < CFN2e-^daN)
for all w G R.
Proof. Theorem 3.2 is a restatement of Lemma 3.6, once one observes
rkh/kh Sij,h)it)dt.
-Nh
A table of aq, q = 1.100, is found on p. 175 of [4]. Also, aq = Si(gîr)/•&■, an
asymptotic expansion for which is given on p. 244 of [1]; we are told this expansion
gives 20 significant figures for q > 20.
Theorem 3.2 is used in conjunction with conformai mapping techniques (ex-
plained in [4, Section 4], and the references therein) to evaluate singular integrals of
the form f¡¡ fit)dt, where a or A or both a and b are finite, and / possibly has
singularities at a or b. We consider here (a, b) = (0,1), and the conformai maps <p
and xp defined in (2.8) and (2.10). From Theorem 3.2 and considerations in Section 2
of this paper, we obtain
Theorem 3.3. Suppose (as in (2.8)-(2.13)) that f E 5(<t)¿), and suppose
(3.22) \f(x)\< C\xf (1 -xf all xEiO, 1),
for some C and ß > 1. Set a = ß + 1, and set h = iird/aN)x/2. Define
(3.21)'
,ajh
FNi*) f(Xj)(Xj)(l - Xj)ae
(eaJh + 1)
S(k,h)(cp(x)) +
Mf) i°N +j >k-j)
xa + (1 - x)a!N(f)
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
568 RALPH BAKER KEARFOTT
where INif) = hlNk=_Nfixk)ixk)il - xk) and where Xj = xpijh) = eJh/il + e'h).
Then, for x G (0,1),
(3-23) \ffit)dt-FNix) CFNle20-(vdaN),/2
for some C dependent on f but independent of N. If /(z) = 0(\zf) as z -» 0, z G C,
then f E Bi%), with d = v/2.
Note that (3.22) is equivalent to
|/(^(w))^(w)[l -xp(w)]\ = \f(x)x(l + x) |« ce"* for w GR.
Theorem 3.3. is otherwise obtained from Theorem 3.2 by making the transformation
(3.24) f fit) dt = rx)fixpiu))xpiu)[l - xPiu)] du,J0 J-x
then applying Theorem 3.2 to the right member of (3.24).
To comment on the power of the approximations in Theorem 3.2 and Theorem
3.3, we note that it is unnecessary to know precisely what the order a is; in
particular, if ¡fit) |< ce~°" and 0 < a' < a, then we may replace a by a' in any of the
formulas. Note that the convergence order is not preserved in this way when we use,
say, Newton-Cotes or Gauss type formulas derived with the method of moments for
weighted integrands of the form xapix), p a polynomial. Also, sine function
approximations are optimal, as explained in [3].
4. Experimental Results. To empirically test formula (3.21)', we tried it for various
TV on the following functions:
(i)/,(x) = |x"2/3;
(ii)/2(x) = f*'/3;
(iii)/3(x) = ¿[x~2/3 + (1 - *r2/3]; and
(iv)/4(*) = ¿[*-9 + (i7*r7]-In each case, the difference FNix) — /0A /(/) dt was computed at x = Ai, i = 1,..., 10,
for TV = 4,8,16, and 32, corresponding to 9,17,33, and 65 evaluations of /,
respectively. These results were compared to the errors Ffiix) — ¡q fit)dt, where
Ffiix) is the result of applying a 2TV-point Gauss-Legendre formula to compute
fo f(t)dt, for 27V = 8,16, and 32; F^ix) was defined to be the result of applying
the composite 32-point Gauss-Legendre formula with 2 subdivisions. These particu-
lar formulas were chosen for comparison since the Gauss formulas are commonly
thought to be generally accurate, and since the 8-, 16-, and 32-point formulas were
closest in number of points to 9,17, and 33 of those formulas already programmed
on our system.
The functions/, / = 1,... ,4, are normalized so that j0x fit)dt = 1. The functions
have singularities at x = 0 and x = 1 of varying orders; the singularities of/3 are the
same order at x = 0 and x = 1, while the singularities of /, /' ¥= 3, differ in order at
x = 0 and x = 1.
Four tables appear, corresponding to each of the four functions. Table 3 has three
parts, corresponding to various estimated orders ß (cf. (3.22)).
The sine function approximations compare favorably to the corresponding Gauss
formulas, except for/2(x). We postulate that the worse the singularity, the better the
merit of the sine function approximation.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
A SINC APPROXIMATION FOR THE INDEFINITE INTEGRAL 569
KÍMMK1KIKIMK1MIO
ooooooooooI I I I I I I 1 I I
rsjooooooooooN"ïmK1fOr\jrOr**>r*rr""IKÏKl
II oo-ooo-oo-o-o-oC ■ ■••*• t * ■ *
ooooooooooI I I ! I t I I I I
ooooooooooI I I I I t I I I I
^OOOOOOOOOO
NI S00Nt-0O0O«-0O(\Jr-ii Kifnio>ïPOM^ro<iMC • • i • • t • ■ ■ i
ooooooooooI I I I I I I I I I
coooooooooo•- I I I I I t I 1 I IOOOOOOOOOOO
Q.f\JOOf\ll/^00*-KlLAOOOI f\JfvjKIKlM>i>T^Nf^
^ t t ••••••• 1
■cooooooooooI I I I I I I I I I
c oooooooooo■r I t I I t I I I I IOOOOOOOOOOO
Q.»**-Oi/>»->Or\tO*mr\jI N*r-r-f\jr\JWMiJir>
-r • • • * t * • • • .•OOOOOOOOOOO
r\irNjrNjf\jr\jr\jr\jr\jr\Jr\j
OOOOOOOOOOI I I I I I I I I I
-OOOOOOOOOOO
i-0«-00*0»-00k*-0
II ir\i/Mn<jiAtfiin-JtMTic • • • •.
ooooooooooI I 1 I I I I I t I
ooooooooooI I I t I I I I I I
OOOOOOOOOOO*— •jinomvfiom'ONoIt f\JMf\jrM(Nji\jMf\ir»jtp
C • • t t.ooooooooooI I I I I I I I I I
COOOOOOCJOOO■♦- I I 1 I I I I I I IOOOOOOOOOOO
l fMn^vj*jiAintni/ivOr\i • • . . • t • • • tOOOOOOOOOOO
I I I I I
*-> OirnnirnninuMnj>ft. OOOOOOOOOO
— I I I I I I I I I IOOOOOOOOOOO
I "Of-rsjKlmOoOO*-*— «-r\j ••••■■••••KIOOOOOOOOOO
ooooooooooI I I I I I I I I I
ooooooooooc0r\j»-Of-^-r\if\jr-O*-II r\jf\jf\ir\jf\jr\ir\jf\tf\jr\jc.•••••
ooooooooooI I I I I I I I I I
coooooooooo-•- I I I I I I I t I IOOOOOOOOOOOû>i>0«jOOOtnWM^
^tn-ûr-NcooocDOOvO
w
<f-
»-ooooooooooI I I I I I I I I I
KlfnroKlKI"iK1"lKirNJ
ooooooooooI I I I I I I I I I
oooooooooo
C • • • • t.ooooooooooI I I I I I I I I I
c oooooooooo— I I I I I I I I I IOOOOOOOOOOOttNKKlfOryOOOO'OI roO>i-MfO>íiríu^>oco
O t • ».• iV oooooooooo
oooooooooo<r I I I I I 1 I I ! III ooooooooooC'OrgiMtrtNOO1 -Of— it*
i/MAiftintruni^ifii^tri
ooooooooooI I I I I I I t I I
r\jrMf\ji\jf\jr\jr\jr\jr\i»-0 ooooooooo1 I 1 I I I I I I »
ooooooooooOOO^-OSf^NeOOr\ifOr\jr\jr\jfOKir\jf\jr\j
»-«-»-0000000cooooooooooC| I I + + + + + + +
■r~ OOOOOOOOOO
Of>Nf>r'Njr\J|<TÍ>í^
I.«oooooooooo
I I I I I I I I I I
V -J'-í'r^lfOKlrOKlrOr'l+*ooooooooocC ( I I I I ( I I I I
-r- OOOOOOOOOO
0(\Jlfl>íííf>«í(MArNQ. <\t ir* o r- t- r\j r\j n «j >jI.
cooooooooooo
r-tvJMsnn-ONWOO(■••••••»••
ooaoooaoo«-
p-ryMsTuTONOOOa
ooooooooo»-
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
RALPH BAKER KEARFOTT
i i i i i i i t i ir\jooOQOO oooofi M «- r\i »- «— r\ji— r\i «— kiII KlKlMKlMMKlrOM'O
C •.oooooooooo
f\jr\jr\jf\if\jrur\jr\jf\jr\j
ooooooooooI I I I I I I I 1 t
ooooooooooI I I I I I I
ooooooooooI I I I I I I I I I
rvjooooooooooKiroi/Mr>ini\jro>i>Of\iroii T-NNTNsrooin>Ofvjtoc •••.
oooooooooo
coooooooooo-r- I I I | I I I I I IOOOOOOOOOOOO.*- -4 C00a>-O<-r\jrO00| *— *— T-t— *— l\l IM IM f\J vt
coooooooooo— I I I I I I I I I I
OOOOOOOOOOOQ.T- vi-OOOO-Oi-fvKleOI «—*—«—*— «— f\J f\J f\l f\J sf
sí • • • «.
■cooooooooooI I I I I I I I I I
coooooooooo■•- I I I I I I I I I I
OOOOOOOOOOOQt— »J,-0000*O»-r\jKï00I r-T-r*T-r-f\|f\jf\JfM*J
«J • •••••••••
-OOOOOOOOOOOI \ ■ I I I t 1 1 I
r\jr\¡r\jt\¡t\jr\ir\ir\ir\ii\j
OOOOOOOOOOI I I I I I I I I I
COOOOOOOOOOt— -*T>Ovr>««TU-lCmsJ--00
II rur\jr\(f\jr\jr\jf\jr\if\iinc.• »
ooooooooooI I I I I I I I I I
r\jr\jr\jr\jr\jfNjr\jfNjr\j»—
ooooooooooI I I I I I I I I I
COOOOOOOOOO*— oCCCCCC-OCi/1H rs.rs.rs.r^-r-.r^rs.r^rs.T-
c • • • t t • • • * •ooooo ooooo
Klsr^lOWKIfOlOflfVJ
ooooooooooI I I I I ! I I I I
coooooooooo*— st iMin st r vOM«. M Kl
II N»nsTKlf-i/>'Or-inKl
C %-#••*••!••oooooooooo
COOOOOOOOOO-*- I I I I t I I I I IOOOOOOOOOOOQs'COOMs'-ONX^'-I i— *— r\jr\jf\jrvir\jr\jr\|sO
r\ji**««««a*f
roOOOoOO OOOOI I I I I I I I I I
COOOOOOOOOO• I- I I I I I I I t I I
OOOOOOOOOOO
I r-r-rypjrMrgfvjrvjfvj^
fM •■■•••••• •
KiOOOOOOOOOO
COOOOOOOOOO■"- t I I I I I I I I IOOOOOOOOOOOQ\toOO'M^<iMO(>'-I f-»-t\jrMrvif\jf\JiMf\J^
rJ t > t t , . • • i »
OOOOOOOOOOOI I I I t I I I I I
ooooooooooI I I I I I I 1 I I
oooooooooo(t-t-O^O**-«- *-OOOi-
t— t— O* O* «— «- T-t—t>r\j
ooooooooooI I I I I I I I I I
OOOOOOOOOOI I I I I I I I I I
oooooooooo00>T sT M M si st sT "1 \T \
II rvjrsir\ifNi'NJ<,sjrsjr\jrsj'sT
£»«••••••••ooooooooooI I 1 I I I I I I I
oooooooooo(0lTiO>0»T-lAsTrsir\M
II i\if>T-'01MMiOfM>Oinc ••••••••• •
oooooooooo
coooooooooof I I I I I I I I I IOOOOOOOOOOO0.r\jco<\Ju-iooor\jsjciA| r\jf\)MfO'»lsTNisT^O
r-0000000000
coooooooooo■r- | I | I I I I I I \OOOOOOOOOOOO. r\j ooruu-ioOO'Msj- Cu^| IMr\iMrAmNTsfsTsrr>
c.■ • ■•-OOOOOOOOOO
COOOOOOOOOO-1111(11111OOOOOOOOOOOÛNOONtnoOOfysT'ÛlAI f\J(\JMMrOsTststsTt>
c • • • •.•-oooooooooo
oooooooooo* I I I t I I I I I 'II ooooooooooCOl^stl^NOr-Oi/I^
Kif\jiMrvjf\jrMroKi(M»n
■ •••tftOfl«oooooooooo
oooooooooo•* t I -I f I I t t I +II ooooooooooCKlOT>0"-'Mriji-l>0
mi/lsTsT^w^w^irtsTr-
oooooooooo
r\jrsir\jrsir\jf\if\jr\jf\jr\i
oooooooooor I I t ! I 1 f ( t I: OOOOOOOOOO
:•— occtniniriO'OOO'•-^stT-MOO-OfON
■ ••«•••«It
oooooooooo
^ooooooooooC I I I I I I I I I -»■
-r- OOOOOOOOOO
OstKlOl^^'MOO-'XJtAQro vT i/"> in ui O c O r«- «—
I • • • • •.oooooooooooo
*-<ooooooooooC I I I i i I i i i +■r- o ooooooooo0«J'K1oir»0,-r\jC0'-<"\JmQ. ki -sTir»ir»i/^cccr~. t—
I ».oooooooooooo
I I I I I I I I I I
cooooooooooC I | I I I | I I | ♦
.,- OOOOOOOOOO
o st Kl u in rj. im <ir> r\ii/>o.Ki^j'v"»u"*u"icO'Ors-«—
*— rNjKi-^u-NC^-OOO^O
OOOOOOOOOt-
t— rMKIsTiriONWOO
ooooooooo»-
r-MKIvtlfl^NOOO'O
ooooooooo»-
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
A SINC APPROXIMATION FOR THE INDEFINITE INTEGRAL
ooooooooooI I I I I I I I
oooooooooo
^oooooooooocoooooooooo■»-♦ + + + + + + + + ♦OOOOOOOOOOOo, m c fs. ooo^o-oOí-fMI MfMIMrMrM'VKlKI'OfO
s* •••••••■••
cooooooooooI I t \ I I I I I I
ooooooooooI I I I I I I I I I
oooooooooorMf-Ot-T-rgryf\iT-rosTststststsTststsflA
• ••*>••••••
ooooooooooI I I I I I I I I I
H
I
+
OOOOOOOOOOOaocoo-0'-i-iNMKii/\| Mr\j(\j|OKlfAKlKiKlKl
r\)t««*«**f*tKIOOOOOOOOOO
I I I I I I I I I I
ooooooooooI I I I I I I I I +
OOOOOOOOOO
OOOr\-fs-o-»— rusí-sl-i-o
c.• • • •ooooooooooI I I I I I » I I I
<
H
coooooooooocoooooooooo.-+ + +++ + + + + +OOOOOOOOOOOO. O r\j >J in^-ONN\r-| KlKlKIKTOKIKIMKlst
•o • • • ».»-OOOOOOOOOO
OOOOOOOOOOOOOOOOOOOO
OOOOOOOOOO
oooooooooocooooooooooC+ + + + + + + + + +
■r- OOOOOOOOOO
OsTrs.OOO-Or-rvlNr'lOO
QK1 rOKlKIst«JsTs»stst
I.OOOOOOOOOOOO
I I I I I I I I I I
»~r\jKisju-tsOr^-0000
OOOOOOOOOt-
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
572 RALPH BAKER KEARFOTT
For Table 1, ß = -2/3 (a = 1/3) was used, while ß = 1/3 in Table 2. The values
ß = -2/3, ß = -.467, and ß = -.867 were used in Tables 3a, 3b, and 3c, respec-
tively, while the value ß = -.9 was used in Table 4. From these and other
experiments, it appears it is better to underestimate ß than to overestimate it.
In all cases, d was taken to be 7r/2.
Stenger has proven a formula analogous to (3.5)', namely
(4.1) f git)dt = h 2 2 ok..,gilh)Sik, h)ix) + OiNe-^d^/2),-°° k = -N l=-N
using formulas (2.17), (3.7), and (3.11) in [4] and the uniqueness of the coefficients in
the sine function expansion [5]. (This formula appears without proof as formula
(3.36) in [4]*.) It is then possible, using the same argument as in the proof of Lemma
3.6, to obtain an analogue to Lemma 3.6 from (4.1). This analogue was also tried in
the same cases as above; in all cases, the tables of errors were the same to at least
two significant digits. Since (4.1) is more elegant and requires less additions, it is
perhaps to be preferred.
The formula resulting from (3.21) when IN is replaced by Ix was also tried. From
the resulting error tables, it appeared that the approximation for Ix accounted for
approximately half of the total error. Thus, it seems unnecessary to use a separate
stepsize h in computing 1N, as suggested by (2.7). Also, taking d other than 77/2 in no
case improved the performance.
The computations were done on a Honeywell 6880 (Multics system) in double
precision (63-bit mantissas). To get values at x = 1, x = 1 — 2"62 was used.
Acknowledgements. I wish to thank Professor Frank Stenger for introducing me to
these approximation techniques. I also wish to thank the referee for his careful
reading and for his suggestions in making the paper more comprehensible and
useful.
Department of Mathematics and Statistics
University of Southwestern Louisiana
Lafayette, Louisiana 70504
1. M. Abramowitz & I. A. Stegun, Editors, Handbook of Mathematical Functions, Dover, New York,
1970.2. F. Stenger, "Approximations via Whittaker's cardinal function," J. Approx. Theory, v. 17, 1976,
pp. 222-240.3. F. Stenger, "Optimal convergence of minimum norm approximations in Hp", Numer. Math., v. 29,
1978, pp. 345-362.4. F. Stenger, "Numerical methods based on Whittaker cardinal, or sine functions," SI A M Rev., v.
23, 1981, pp. 165-224.5. F. Stenger, private communication.
* Formula (3.36) in [4] was attributed to Kearfott, Sikorski, and Stenger; that paper never existed. Also,
the derived formula (4.58) in [4] was mistakenly attributed to John Lund; that formula corresponds to
(3.23) of this paper.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use