An Elementary Proof of the Error Estimates in Simpson's Rule

Author(s): D. D. Hai and R. C. Smith

Source: Mathematics Magazine, Vol. 81, No. 4 (Oct., 2008), pp. 295-300
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/27643127
Accessed: 06-12-2015 23:10 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/27643127?seq=1&cid=pdf-reference#references_tab_contents

You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to Mathematics
Magazine.

http://www.jstor.org

This content downloaded from 129.81.226.78 on Sun, 06 Dec 2015 23:10:27 UTC
All use subject to JSTOR Terms and Conditions
VOL. 81, NO. 4, OCTOBER 2008 295

ProofWithout Words: Tangent of the Sum

tan/? C tana B

AC = 1

BC = tana

DC = tana tan?

EC =
tan?
DF = EB
tana + tan?
tan(a + ?)
1? tana tan?

?Sidney H. Kung
Cupertino, CA 95014-2292
sidneykung @ yahoo.com

An Elementary Proof of the Error Estimates

in Simpson's Rule
D. D. HAI
Mississippi State University
Mississippi State, MS 39762

R. C. SMITH
Mississippi State University
Mississippi State, MS 39762

Recently, Cruz-Uribe and Neugebauer [1] gave an elementary proof of error estimates
for the Trapezoidal rule, which based on integration by parts "backwards." However,
theywere unable to extend theirmethod to the error estimates in Simpson's rule. In

This content downloaded from 129.81.226.78 on Sun, 06 Dec 2015 23:10:27 UTC
All use subject to JSTOR Terms and Conditions
296 MATHEMATICS MAGAZINE

this note, we shall provide a new and elementary proof of error estimates in Simpson's
rule for functions having four or three derivatives. Our approach also depends upon
integration by parts and thus answers an open question posed in [1]. Note that in a more
advanced paper [2], the authors derived error estimates inTrapezoidal and Simpson's
rule for functions possessing just one or two derivatives. For other elementary proofs
of error estimates in Simpson's rule, we refer toTaiman [4] and the references therein.
To set our notation we let / : [a, b] ? M and for n e N, we put Xj = a +
j(h2~a)
for j = 0, 1, ... , 2n, and put

S2n= [/(*2*-2)+ 4/(*2*_!)+ f(x2k)] .

b-^ V

If / is four times differentiable on [a, b] and K > 0 is a constant such that <
|/(4)(*) |
K for all x e [a,b], then a standard result quoted in nearly every elementary calculus
book is the error estimate
b
- K(b-a)5
f(x)dx S2n -
180(2rc)4
1/
For n > 1, S2n is called the composite Simpson's Rule inmore advanced texts.We do
themajority of our work in the situation where n = 1, h > 0, [a, b] ? \?h, h] and
-
S2 = | [/(-A) + 4/(0) + /(A)]. PutE2 = ?h_h
f{x) dx S2.
We are interested in estimates on E2 for functions / that are at least three times
differentiable on [?A, A] and we do some preliminary calculations towrite E2 as an
integral formula. In all subsequent integration by parts calculations throughout this
paper, constants are chosen to make the polynomial factor have a zero at A.
= allows us to replace =
Putting g(x) f(x) + f(?x) f_h f(x)dx f0 g(x)dx.
Now integrate by parts, using u = dv = du = and v = x ?
g(x), dx, g\x) dx, y to
get

= g(x)dx = -g(h) + lx- ? \g/(x)dx.

J(x)dx j ?g(0)-j
second integration by parts, using u = dv = ?
A g'(x), (jc y) dx, du = g"{x)dx,
and v = y ? + y gives us
^f-

- Ch A rh ?x1 2hx -h2\

f{x) dx = [/(-A) + 4/(0) + /(A)] + I- -? + g"ix) dx.
J ^ J J

Consequently,

r (x2
,2 2hx
ihv
h2\
h2
?

We startwith themost familiar estimate:

THEOREM 1. Let f be four times differentiable on [?A, A] and suppose K > 0 is

instant such that < K <
|/(4)C*)| for all x e [a, b]. Then \E2\ ^|jp
Proof. From (1) and integrating by parts again, we obtain

(h(x2 2hx h2\ ? x, [hfx3 hx2 h2x\ ?,

(2)

This content downloaded from 129.81.226.78 on Sun, 06 Dec 2015 23:10:27 UTC
All use subject to JSTOR Terms and Conditions
VOL. 81, NO. 4, OCTOBER 2008 297
x{x-h)2 > hx1 _
h2x -
Notethatg"'(0)=0, \g^(x)\ <2tfforalljt e [0,A],and 6 3 , 6 6
0. By theMean Value Theorem, for each x e (0, A], there is a point ^ in (0, x) such
that g'"(x) = xg(4)(sx). Hence

-1 fV? -? ? ^ix^wfe)?/x <2# ?V? -? ?^ xdx

Kh5
~90~

In both the statement and proof of the next result,we make use of real analysis. The
particular notions and results from real analysis thatwe need are sometimes called
the Fundamental Theorem of the Calculus for Lebesgue Integrals. All thatwe need
can be found in pages 108-110 of [3].We briefly summarize as follows: g is abso
lutely continuous on [a, b] if and only if g! exists almost everywhere on [a, b] and
=
g' is Lebesgue integrable on [a, b]. In this case, g(x) g(a) + j* gf(t)dt for all
x e [a, b]. We let Ll(a, b) denote the collection of Lebesgue integrable functions on
and for g e Ll we =
[a,b] (a, b), put \\g\\Li(a,b) f^\g(x)\dx.
THEOREM 2. Let f be three times differentiate on [?h, A] and suppose f" is
<
absolutely continuous on [?h,h]. Then \E2\ j^Wf^Woi-h^y

Proof. Using (2) and integration by parts again give

?h(xA
? - ? hx3 h2x2- ? A4\ (4)
E7= I I +
+ -^- Ig(4)(x)dx. (3)
"Jo V24~~<r l2"~72/
Put0(x) = f^-^ + ^?-^and note that0(A) = 0,
?
=
x3 hx2 h2x x(x h)2 >
*'(*> 0 on [0, h].
T --r +-?

Thus hi < < 0, and so

72 <P(x)

=
\gw(x)\dx
J \<P(x)gm(x)\dx<^^ ^JJfw(x)\dx

?l/Wl
Ll(~h,h)'

THEOREM 3. Let f be three times differentiate on [?h, h] and suppose there is a

> 0 such that < <
constant M \f'"(x)\ M for all x e [-h, A]. Then \E2\ ^p.

Proof. By theMean Value Theorem, for each x e [0, A], there exists a point sx
between x and A/3 such that

=
g"(x) g"(h/3) + (x- h/3)g'"(sx).

2hx =
Since f 0 and \g'"{x)\ < 2M for all x e [0, h], it follows that
?(' tf\dx

Ifh (x2 2hx h2\ ?t xJ

m

Ifh (x2 2hx h2\ ( h\ mi xJ

(s*)dx
Hi (t-?+v{x--3)8

This content downloaded from 129.81.226.78 on Sun, 06 Dec 2015 23:10:27 UTC
All use subject to JSTOR Terms and Conditions
 298 MATHEMATICS MAGAZINE

?l?
lo
i3x-h)2ix-h)g'"isx)dx
MO

2M 9 - , MA4
<- fh -
A)2 (A jc)djc=
// Qx
18 /o
Jo
(3x
36
We give the error estimates for theComposite Simpson's Rule as a corollary:
COROLLARY 4. Let f : [a,b] ?> /? ???continuous, let S2n be defined as above, let
n e N and letE2n = ?
fa fix)dx S2n. Then thefollowing hold:

(i) (/"/ isfour times differentiable on [a, b] and K > 0 is a constant such that
< x <
|/<4)(*)| K for all e [a,b] then\E2n\ fg=f?;
(ii) iff is three times differentiable on [a, b] and f" is absolutely continuous on

{aMthen\E2n\<^\\fW\\LHa,by
(iii) is three times differentiable on [a, b] and M > 0 is a constant such that
iff
<
\f'"ix)\ M for all x e [a, b] then \E2n\<
**g?.
Proof. For themoment, fix k e {1,... ,n). Let A= and put gix) = fix +
-^
for ?A < x < A. Note that
x2ic-\)

2k
-
f{x) dx ?^- + 4/(X2t-l)+ /(**)]
[f(X2k-2)
I

= + g(h)].
j g(x)dx-j[g(-h)+4g(fl)
If / is four times differentiable on [a,b] and K > 0 is a constant such that
< ^ for all x e
|/<4)(*)| [a, b] then g is four times differentiable on [?h, h] and
= < K for aux e [-A,
\gW(x)\ |/<4)(*+*2*-i)| A]. By Theorem 1,

=
1/ *HxUx-?li(-h)+4m+m]\^ >^
If / is three times differentiable on [a, b] and f" is absolutely continuous on [a, b]
then g is three times differentiable on [?A, A] and g'" is absolutely continuous on
[?A, A]. Moreover,

\r\^\dx=rJx2k-2 \^\x)\dx,
\-^\LH_hM)=?
J-h

and so by Theorem 2,
"
A I
?(*) dx-- [?(-A) + 4?(0) + ?(A)]
1/
?4
-<^! ?-mil (?'-a)4 r2t
72 11,(4)1.
II* IIlI(-m)=(^^P /M !/
72(2n)4 '*2*-2
|/(4)(;c)kjc
W|^'

If / is three times differentiable on [a, b] and M > 0 is a constant such that <
|/'"(*) |
M for all x g [a, b] then g is three times differentiable on [?A, A] and |g"'(jt)| < ^
for all x e [?A, A] so by Theorem 3,

MA4 -
Mib a)4
gix) dx-\ [gi-h) + 4gi0) + ?(A)] 36
1/ 36(2rc)4

This content downloaded from 129.81.226.78 on Sun, 06 Dec 2015 23:10:27 UTC
All use subject to JSTOR Terms and Conditions
VOL. 81, NO. 4, OCTOBER 2008 299
The estimates in (i) and (iii) are now a matter of summing over k and using the triangle
inequality:
In (i) we have

n r*2k ?
b a
^2n\ f(x)dx 6n
[/(*2*-2)+4/(*2*-l) + /(*2*)]
k=\ x2k-2

n I /.x2k

/(*) dx - ?^ [f(x2k.2) + 4f(X2k-i) + f(x2k)]

k=\
on
A \Jx:

< K(b-a)5 K(b-a)5

n '
90(2n)5 180(2n)4

while in (iii) we have

- -
M(b a)4 M(b a)4
<n
\E2n\
36(2n)4 72(2n)3
The estimate in (ii) is also obtained by summing over k but we make use of

i/(4)u,,=/V4)(*)id*=?Ja k=l
rJx2k-2 \f\x)\dx
to see that

Eln\<y(b a\ fX2k\f4\X)\dx
;| (^-^)4||f(4)||
" HLHa,b)m
-f^72(2n)4 JX2k_2lJ 72(2n)4

Remarks, (i) In Theorem 1, ifwe assume /(4) is continuous on [?A, A], thenwe can
avoid the use of theMean Value Theorem and simply use the formula (3) to obtain

h '-*
dX= Kh5
2v2
hZX h4\
\E2 hx^
24 9 12 ^
?r(? ~V2

A similar remark can be made about Theorem 3.

(ii) An even better approach is to use theLebesgue integral instead of theRiemann
integral. In this situation, it is enough to assume /(3) is absolutely continuous and
f(4) is essentially bounded. Moreover, K can be replaced by the essential supremum
of |/(4)| and we don't need a uniform bound on /(4). The major disadvantage of this
approach (and our Theorem 2) is that it requires more analysis knowledge thanwe find
with typical undergraduates.
(iii)We note that all of our error estimates are sharp. The central result in [4] gives
an accessible argument thatTheorem 1 is sharp. To see thatTheorem 2 is sharp, con
sider the situation with 0 < s < A and

0 forje < ?s,

feW = { 72+ 15+ 8 + 9+ 24e 360e3

IOr 1*1- ?>
2e?x i 2x
15 ^ 9 for x > s.

It is tedious but straightforward to verify that fe is four times continuously differen

=
tiableon [-A, A] and f\ \f?(4)(x)\dx and - < 0(s2).
\E2\
? ? ?

This content downloaded from 129.81.226.78 on Sun, 06 Dec 2015 23:10:27 UTC
All use subject to JSTOR Terms and Conditions
300 MATHEMATICS MAGAZINE

Finally, to see thatTheorem 3 is sharp,we again consider the case when 0 < e < A,
but this timewe put

for 0 < x < s,

=
fsix) lex1 i e2jc ~ 2e
< x < A,
3 + ^ is" fore

and extend f? to be an even function. Another straightforward but tedious argument

shows thatf? is three times continuously differentiable and < 2 for all x e
|/"' (x)\ ,4
?
[?A, A]. In this case M 2 and our estimate in Theorem 3 becomes -^; a direct cal
-
culation shows
y^ \E2\ Ois2). Note that in [4], Taiman shows that \E2\< 2^f-.
<
This is similar in form to our result but not sharp as Taiman himself points out.

Acknowledgements. The authors wish to thank the referees for pointing out some errors in the examples in an
earlier version of this paper and for their suggestions for improving our exposition.

REFERENCES
1. D. Cruz-Uribe and C. J.Neugebauer, An elementary proof of error estimates for the trapezoidal rule, this
Magazine 76 (2003) 303-306.
2. D. Cruz-Uribe and C. J.Neugebauer, Sharp error bounds for the trapezoidal rule and Simpson's rule, J. Ineq.
Pure Appl. Math 3(4) (2002) article 49.
3. H. L. Royden, Real Analysis, 3rd Ed., Macmillan, New York, 1988.
4. L. A. Taiman, Simpson's rule is exact for quintics, Amer. Math. Monthly 113 (2006) 144-155.

The Class of Heron Triangles

RAYMONDA. BEAUREGARD
University of Rhode Island
Kingston, Rl 02881
[email protected]

A Heron triangle is one with rational sides and rational area. A glance at sources
ranging from Dickson's wonderful compilation [1] to themodern Wolfram website [2]
indicates that such triangles have not only been but will continue to be a fascination. If
a Heron triangle is scaled up using the lowest common multiple of the denominators
of the sidelengths, then the triangle is similar to one having integer sides and integer
area. Three parameters m,n,k can then be that generate all such Heron
integer given
triangles: the sides have the form (see [2])
OOOO ? o
n(m +k), m(n + k ), (m + n)(mn k).

In this note we startwith a Heron triangle and scale it (usually down) to produce a
Heron triangle with altitude 2.We are then able to obtain two rational parameters that
generate this family of Heron triangles.
THEOREM. After scaling, everyHeron triangle has theform of the right-most tri
angle in thefigure below, for rationals r and s with r, s > 1.

Proof. Any triangle can be oriented so that ithas the configuration of the first tri
angle in the figure. Ifwe assume that it is a Heron triangle and ifwe multiply its sides
A, B, C and its (rational) altitude A by the scaling factor 2/A, we obtain the second
triangle in the figure with rational sides a, b, c and altitude 2.

This content downloaded from 129.81.226.78 on Sun, 06 Dec 2015 23:10:27 UTC
All use subject to JSTOR Terms and Conditions