TSUKUBA J. MATH.

Vol. 33 No. 2 (2009), 281–297

CONVERGENCE RATES OF APPROXIMATE SUMS

OF THE AREAS OF SURFACES OF REVOLUTION

By
Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

Abstract. We represent the convergence rates of approximate sums

of the areas of surfaces of revolution as limits of their expanded error
terms and estimate them. In the case of convex surfaces of revolution
we represent the convergence rates of them by the integral of certain
functions.

1. Introduction

Let ½a; b
be a bounded closed interval and f be a function of class C 1

defined on ½a; b
. We assume f > 0 and consider the surface of revolution defined
by f :

fðx; f ðxÞ cos y; f ðxÞ sin yÞ j a a x a b; 0 a y a 2pg:

The area S of this surface of revolution is given by

ðb
S ¼ 2p f ðxÞf f 0 ðxÞ 2 þ 1g 1=2 dx:
a

We take an n-division Dn of ½a; b
defined by

Dn : a ¼ s0 a s1 a    a sn1 a sn ¼ b:

We join ðsi1 ; f ðsi1 ÞÞ and ðsi ; f ðsi ÞÞ by segment and revolve the obtained
polygon around the x-axis. At this point we obtain a union of frustums, whose

2000 Mathematics Subject Classification. 53A05 (Primary), 41A21 (Secondary).

Key words and phrases. convergence rate; approximate sum; area; surface of revolution.
The second author was partly supported by the Grant-in-Aid for Science Research (C) 2009 (No.
21540063), Japan Society for the Promotion of Science.
Received April 20, 2009.
Revised June 29, 2009.
282 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

area SðDn Þ is given by

Xn
f ðsi1 Þ þ f ðsi Þ
SðDn Þ ¼ 2p fð f ðsi Þ  f ðsi1 ÞÞ 2 þ ðsi  si1 Þ 2 g 1=2 :
i¼1
2

We regard SðDn Þ as an approximate sum of the area S.

In the present paper we consider the limits of the expanded error terms

ð *Þ n 2 jS  SðDn Þj and n 2 ðS  SðDn ÞÞ;

which represent the convergence rates of the approximate sums SðDn Þ of S.

Our purpose is to estimate the limits of (*), stated in Theorems 1.2, 1.3, and
1.4.
In the case of the lengths of curves Gleason [2] has treated convergence rates
of some approximate sums of the lengths of curves and obtained the following
theorem.

Theorem 1.1 (Gleason). Let A be a curve in euclidean space of class C 2 and

length L. For each positive integer n, let Pn be the longest polygon of n edges
properly inscribed in A. Then

ð 3
1
lim n 2 ðL  LðPn ÞÞ ¼ k 2=3 ds ;
n!y 24 A

where k is the curvature of A.

In this case for any polygon P inscribed in A we have L b LðPÞ, however in

our case we do not know which of S and SðDÞ is greater for a division D of ½a; b
.
So we cannot use Theorem 1.1 directly to the case of the areas of surfaces of
revolution, but some of arguments and a lemma in Gleason [2] are still useful in
our case.
Schwarz showed an example of polyhedra inscribed in a cylinder whose areas
did not converge to the area of the cylinder in [3], which is now called the lantern
of Schwarz. So we have to be careful in treating approximate sums of the areas
of surfaces.
Approximate sums of Riemann integrals are the most fundamental ap-
proximate sums. Chui [1] and the second named author [4] obtained some results
on the approximate sums of Riemann integrals. It is possible to apply their results
to the function f ðxÞf f 0 ðxÞ 2 þ 1g 1=2 of our case, however with the approximate
sums consisting the derivatives of f ðxÞ. Since our approximate sums can be ob-
 Approximate sums of the areas of surfaces of revolution 283

tained directly from the surface of f without the derivatives, our approximation
is useful.
For an n-division Dn of ½a; b
, we define dðDn Þ by

dðDn Þ ¼ maxfsi  si1 j 1 a i a ng:

For a sequence fDn gy y

n¼1 of n-divisions of ½a; b
, we define eðfDn gn¼1 Þ by

eðfDn gy
n¼1 Þ ¼ lim sup ndðDn Þ:
n!y

This is useful for estimate of convergence rate of approximate sums.

From now on we assume that the function f is of class C 4 , because we use
the fourth derivative of f in the proofs of Theorems 1.2, 1.3, and 1.4. In order to
state the theorems we define a function j by

1 1
jðxÞ ¼  f 00 ðxÞf f 0 ðxÞ 2 þ 1g1=2 þ f ðxÞ f 00 ðxÞ 2 f f 0 ðxÞ 2 þ 1g3=2 :
2 4

This function j is the coe‰cient of the third term of the local error term, which
we show in Corollary 2.3.

Theorem 1.2. For a sequence fDn gy

n¼1 of n-divisions of ½a; b
, we have

p 3
lim sup n 2 jS  SðDn Þj a eðfDn gy
n¼1 Þ max jjj:
n!y 3 ½a; b

The left hand side of the inequality in the above theorem is a limit of the
expanded error terms mentioned in the abstract. In the case where j is positive
we can get a sharper estimate of expanded error terms for some divisions defined
as follows: The set of all n-divisions of ½a; b
is compact and

D 7! SðDÞ

is continuous, so there exists an n-division Dan at which jS  SðDÞj attains its

minimum. This n-division Da n is optimal for the approximate sum SðDÞ. It may
a
not be unique, but the error term jS  SðDn Þj is unique. Thus we can consider
jS  SðDan Þj.

Theorem 1.3. If j > 0 holds, then we have

ð b 3
2 a p 1=3
lim sup n jS  SðDn Þj a jðxÞ dx :
n!y 3 a
284 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

If the surface of revolution is convex, that is, if f 00 < 0, then j > 0 and
S  SðDÞ b 0 for any division D of ½a; b
. In this case we can represent the limit
of expanded error terms by the same integral in the above theorem.

Theorem 1.4. If the surface of revolution is convex, that is, f 00 < 0, then we
have

ð b 3
p
lim n 2 ðS  SðDa
n ÞÞ ¼ jðxÞ 1=3 dx :
n!y 3 a

The authors are grateful to Kazuyuki Enomoto for directing their attention
to the paper [2] and explaining about it. This was the starting point of this
research. The authors also thank the referee, whose many useful comments
improved the manuscript.

2. Local Error Terms

For a subinterval ½ p; q
H ½a; b
we define the local error term F ðp; qÞ by

ðq
F ðp; qÞ ¼ f ðxÞf f 0 ðxÞ 2 þ 1g 1=2 dx
p

f ð pÞ þ f ðqÞ
 fð f ðqÞ  f ð pÞÞ 2 þ ðq  pÞ 2 g 1=2 :
2
The first term of the right hand side of the definition of F ðp; qÞ multiplied by
2p is equal to the local area of the surface of revolution and the second term
multiplied by 2p is equal to the area of the corresponding frustum. We ex-
press F ð p; qÞ by the linear combination of ðq  pÞ i ð0 a i a 4Þ in Corollary
2.4. Using the expression we will prove our main theorems in Sections 3, 4,
and 5.
In order to make the calculation simpler we change variables and functions
by

u ¼ q  p; gðuÞ ¼ f ðqÞ  f ð pÞ:

The new function g satisfies

gð0Þ ¼ 0; gðiÞ ðuÞ ¼ f ðiÞ ðqÞ ði b 1Þ:

We set I ¼ ½0; b  a
and

hðuÞ ¼ ðgðuÞ 2 þ u 2 Þ 1=2 :

 Approximate sums of the areas of surfaces of revolution 285

I is the range of the variable u. Using h we can write F ðp; qÞ as follows:

ð pþu
1
F ðp; qÞ ¼ f ðtÞð f 0 ðtÞ 2 þ 1Þ 1=2 dt  ð2f ð pÞ þ gðuÞÞhðuÞ:
p 2

By the definition of h, we note that if g is of class C r , then h is also of class C r .

We need the following lemma to calculate the higher derivatives of F .

Lemma 2.1. If g is of class C 4 , then we obtain the following equalities.

h 0 ðuÞ ¼ hðuÞ1 ðgðuÞg 0 ðuÞ þ uÞ ðu > 0Þ;

h 0 ð0Þ ¼ ðg 0 ð0Þ 2 þ 1Þ 1=2 ;

h 00 ðuÞ ¼ hðuÞ1 ðg 0 ðuÞ 2 þ gðuÞg 00 ðuÞ þ 1  h 0 ðuÞ 2 Þ ðu > 0Þ;

h 00 ð0Þ ¼ h 0 ð0Þ1 g 0 ð0Þg 00 ð0Þ;

h 000 ðuÞ ¼ hðuÞ1 ð3g 0 ðuÞg 00 ðuÞ þ gðuÞg 000 ðuÞ  3h 0 ðuÞh 00 ðuÞÞ ðu > 0Þ;

3
h 000 ð0Þ ¼ h 0 ð0Þ3 g 00 ð0Þ 2 þ h 0 ð0Þ1 g 0 ð0Þg 000 ð0Þ;
4
hð4Þ ðuÞ ¼ hðuÞ1 ð3g 00 ðuÞ 2 þ 4g 0 ðuÞg 000 ðuÞ þ gðuÞgð4Þ ðuÞ

 3h 00 ðuÞ 2  4h 0 ðuÞh 000 ðuÞÞ ðu > 0Þ;

3 1
hð4Þ ð0Þ ¼  h 0 ð0Þ5 g 0 ð0Þg 00 ð0Þ 3 þ h 0 ð0Þ3 g 00 ð0Þg 000 ð0Þ þ h 0 ð0Þ1 g 0 ð0Þgð4Þ ð0Þ:
8 2

Proof. By the definition of h we obtain the following equalities.

hðuÞh 0 ðuÞ ¼ gðuÞg 0 ðuÞ þ u ðu A I Þ;

h 0 ðuÞ ¼ hðuÞ1 ðgðuÞg 0 ðuÞ þ uÞ ðu > 0Þ;

h 0 ð0Þ ¼ lim h 0 ðuÞ ¼ ðg 0 ð0Þ 2 þ 1Þ 1=2 :

u!0

Di¤erentiating the first equality of the above we obtain

ð2:1Þ h 0 ðuÞ 2 þ hðuÞh 00 ðuÞ ¼ g 0 ðuÞ 2 þ gðuÞg 00 ðuÞ þ 1

and

h 00 ðuÞ ¼ hðuÞ1 ðg 0 ðuÞ 2 þ gðuÞg 00 ðuÞ þ 1  h 0 ðuÞ 2 Þ ðu > 0Þ:

286 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

Using this we consider the limit

h 00 ð0Þ ¼ lim h 00 ðuÞ ¼ 3h 0 ð0Þ1 g 0 ð0Þg 00 ð0Þ  2h 00 ð0Þ

u!0

and get

h 00 ð0Þ ¼ h 0 ð0Þ1 g 0 ð0Þg 00 ð0Þ ¼ ðg 0 ð0Þ 2 þ 1Þ1=2 g 0 ð0Þg 00 ð0Þ:

Di¤erentiating (2.1) we obtain

ð2:2Þ 3h 0 ðuÞh 00 ðuÞ þ hðuÞh 000 ðuÞ ¼ 3g 0 ðuÞg 00 ðuÞ þ gðuÞg 000 ðuÞ

and

h 000 ðuÞ ¼ hðuÞ1 ð3g 0 ðuÞg 00 ðuÞ þ gðuÞg 000 ðuÞ  3h 0 ðuÞh 00 ðuÞÞ ðu > 0Þ:

Using this we consider the limit

h 000 ð0Þ ¼ lim h 000 ðxÞ

x!0

¼ 3h 0 ð0Þ1 g 00 ð0Þ 2 þ 4h 0 ð0Þ1 g 0 ð0Þg 000 ð0Þ  3h 0 ð0Þ1 h 00 ð0Þ 2  3h 000 ð0Þ

and get

3
h 000 ð0Þ ¼ h 0 ð0Þ3 g 00 ð0Þ 2 þ h 0 ð0Þ1 g 0 ð0Þg 000 ð0Þ:
4

Di¤erentiating (2.2) we obtain

3h 0 ðuÞh 00 ðuÞ þ hðuÞh 000 ðuÞ ¼ 3g 0 ðuÞg 00 ðuÞ þ gðuÞg 000 ðuÞ

and

hð4Þ ðuÞ ¼ hðuÞ1 ð3g 00 ðuÞ 2 þ 4g 0 ðuÞg 000 ðuÞ þ gðuÞgð4Þ ðuÞ  3h 00 ðuÞ 2  4h 0 ðuÞh 000 ðuÞÞ:

Using this we consider the limit

hð4Þ ð0Þ ¼ lim hð4Þ ðuÞ

u!0

¼ 10h 0 ð0Þ1 g 00 ð0Þg 000 ð0Þ þ 5h 0 ð0Þ1 g 0 ð0Þgð4Þ ð0Þ

 10h 0 ð0Þ1 h 00 ð0Þh 000 ð0Þ  4hð4Þ ð0Þ:

and get

3 1
hð4Þ ð0Þ ¼  h 0 ð0Þ5 g 0 ð0Þg 00 ð0Þ 3 þ h 0 ð0Þ3 g 00 ð0Þg 000 ð0Þ þ h 0 ð0Þ1 g 0 ð0Þgð4Þ ð0Þ: r
8 2
 Approximate sums of the areas of surfaces of revolution 287

Lemma 2.2. If f is of class C 4 , then we obtain the following equalities.

 
qF ðp; qÞ  q 2 F ð p; qÞ 
F ðp; pÞ ¼ 0; ¼ 0; ¼ 0;
qq q¼p qq 2 q¼p

q 3 F ðp; qÞ  1 1
¼  h 0 ð0Þ1 g 00 ð0Þ þ f ð pÞh 0 ð0Þ3 g 00 ð0Þ 2 ;
qq 3 q¼p 2 4

q 4 F ð p; qÞ
¼ ðg 0 ðuÞ 2 þ 1Þ 1=2 g 000 ðuÞ þ 3ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞg 00 ðuÞ 2
qq 4

þ 3ðg 0 ðuÞ 2 þ 1Þ3=2 g 0 ðuÞg 00 ðuÞ 2 þ 3ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞ 2 g 000 ðuÞ

þ ð f ð pÞ þ gðuÞÞf3ðg 0 ðuÞ 2 þ 1Þ5=2 g 0 ðuÞg 00 ðuÞ 3

þ 3ðg 0 ðuÞ 2 þ 1Þ3=2 g 00 ðuÞg 000 ðuÞ þ ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞgð4Þ ðuÞg

1
 fgð4Þ ðuÞhðuÞ þ 4g 000 ðuÞh 0 ðuÞ þ 6g 00 ðuÞh 00 ðuÞ þ 4g 0 ðuÞh 000 ðuÞ
2
þ ð2f ð pÞ þ gðuÞÞhð4Þ ðuÞg:

Proof. The definition of F implies F ðp; pÞ ¼ 0. We calculate the higher

derivatives of F .
First we get

qF ðp; qÞ 1 1
¼ ð f ðpÞ þ gðuÞÞðg 0 ðuÞ 2 þ 1Þ 1=2  g 0 ðuÞhðuÞ  ð2f ð pÞ þ gðuÞÞh 0 ðuÞ
qq 2 2

and

qF ðp; qÞ 
¼ f ðpÞðg 0 ð0Þ 2 þ 1Þ 1=2  f ð pÞh 0 ð0Þ ¼ 0:
qq q¼p

Secondly we get

q 2 F ð p; qÞ
¼ g 0 ðuÞðg 0 ðuÞ 2 þ 1Þ 1=2 þ ð f ð pÞ þ gðuÞÞðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞg 00 ðuÞ
qq 2
1
 fg 00 ðuÞhðuÞ þ 2g 0 ðuÞh 0 ðuÞ þ ð2f ðpÞ þ gðuÞÞh 00 ðuÞg
2
and
288 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa


q 2 F ðp; qÞ 
¼ g 0 ð0Þðg 0 ð0Þ 2 þ 1Þ 1=2 þ f ðpÞðg 0 ð0Þ 2 þ 1Þ1=2 g 0 ð0Þg 00 ð0Þ
qq 2 q¼p

 g 0 ð0Þh 0 ð0Þ  f ð pÞh 00 ð0Þ

¼ 0:

Using the equality

fðg 0 ðuÞ 2 þ 1Þ 1=2 g 00 ¼ ðg 0 ðuÞ 2 þ 1Þ3=2 g 00 ðuÞ 2 þ ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞg 000 ðuÞ
and the Leibniz rule we obtain

q 3 F ð p; qÞ
¼ ðg 0 ðuÞ 2 þ 1Þ 1=2 g 00 ðuÞ þ 2ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞ 2 g 00 ðuÞ
qq 3

þ ð f ð pÞ þ gðuÞÞfðg 0 ðuÞ 2 þ 1Þ3=2 g 00 ðuÞ 2 þ ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞg 000 ðuÞg

1
 fg 000 ðuÞhðxÞ þ 3g 00 ðuÞh 0 ðuÞ þ 3g 0 ðuÞh 00 ðuÞ þ ð2f ð pÞ þ gðuÞÞh 000 ðuÞg
2
and

q 3 F ð p; qÞ  1 1
¼  h 0 ð0Þ1 g 00 ð0Þ þ f ð pÞh 0 ð0Þ3 g 00 ð0Þ 2 :
qq 3 q¼p 2 4
Using the equality

fðg 0 ðuÞ 2 þ 1Þ 1=2 g 000 ¼ 3ðg 0 ðuÞ 2 þ 1Þ5=2 g 0 ðuÞg 00 ðuÞ 3 þ 3ðg 0 ðuÞ 2 þ 1Þ3=2 g 00 ðuÞg 000 ðuÞ

þ ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞgð4Þ ðuÞ

and the Leibniz rule we obtain

q 4 F ðp; qÞ
¼ ðg 0 ðuÞ 2 þ 1Þ 1=2 g 000 ðuÞ þ 3ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞg 00 ðuÞ 2
qq 4

þ 3ðg 0 ðuÞ 2 þ 1Þ3=2 g 0 ðuÞg 00 ðuÞ 2 þ 3ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞ 2 g 000 ðuÞ

þ ð f ðpÞ þ gðuÞÞf3ðg 0 ðuÞ 2 þ 1Þ5=2 g 0 ðuÞg 00 ðuÞ 3

þ 3ðg 0 ðuÞ 2 þ 1Þ3=2 g 00 ðuÞg 000 ðuÞ þ ðg 0 ðuÞ 2 þ 1Þ1=2 g 0 ðuÞgð4Þ ðuÞg

1
 fgð4Þ ðuÞhðuÞ þ 4g 000 ðuÞh 0 ðuÞ þ 6g 00 ðuÞh 00 ðuÞ þ 4g 0 ðuÞh 000 ðuÞ
2
þ ð2f ðpÞ þ gðuÞÞhð4Þ ðuÞg: r

Lemmas 2.1 and 2.2 imply the following corollary.

 Approximate sums of the areas of surfaces of revolution 289

Corollary 2.3.

q 3 F ð p; qÞ  1 1
3  ¼  ð f 0 ðpÞ 2 þ 1Þ1=2 f 00 ðpÞ þ f ð pÞð f 0 ðpÞ 2 þ 1Þ3=2 f 00 ð pÞ 2 :
qq q¼p 2 4

The right hand side of the above equality is nothing but the definition of j in
Section 1. This is the reason why we consider the function j.
Using Taylor’s theorem, Lemma 2.2, and Corollary 2.3 we obtain an ex-
pression of F as follows:

Corollary 2.4. For a subinterval ½p; q
H ½a; b
there exists p < r < q with
property that

1 1 q 4 F ð p; qÞ 
F ðp; qÞ ¼ jðpÞðq  pÞ 3 þ ðq  pÞ 4 :
3! 4! qq 4 q¼r

The explicit expression of the fourth derivative of F is given in Lemma 2.2.

3. Proof of Theorem 1.2

Before the proof of Theorem 1.2 we mention several fundamental properties

of eðfDn gy
n¼1 Þ defined in Section 1.

Proposition 3.1. For a sequence fDn gy

n¼1 of n-divisions of ½a; b
, we have

b  a a eðfDn gy
n¼1 Þ a y:

Moreover for any a satisfying b  a a a a y there exists a sequence fDn gy

n¼1 of
y
n-divisions of ½a; b
with property that eðfDn gn¼1 Þ ¼ a.

Proof. For any positive integer n we have ðb  aÞ=n a dðDn Þ, so we get

b  a a eðfDn gy
n¼1 Þ a y.
We take a satisfying b  a a a a y. If a < y, then we define n-divisions Dn
for n > a=ðb  aÞ by

a a i1
s1 ¼ a þ ; si ¼ a þ þ ðb  s1 Þ ð2 a i a n  1Þ:
n n n1

We can see eðfDn gy

n¼1 Þ ¼ a. If a ¼ y, then we define n-divisions Dn by

aþb aþb i1

s1 ¼ ; si ¼ þ ðb  s1 Þ ð2 a i a n  1Þ:
2 2 n1

We can see eðfDn gy

n¼1 Þ ¼ y. r
290 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

Roughly speaking eðfDn gy y

n¼1 Þ measures the di¤erence between fDn gn¼1 and
the sequence of regular n-divisions defined by

i
si ¼ a þ ðb  aÞ ð1 a i a nÞ:
n

Remark 3.2. There exists a sequence fDn gy n¼1 of n-divisions of ½a; b
which
satisfies eðfDn gy
n¼1 Þ ¼ y and lim dðDn Þ ¼ 0. We define n-divisions Dn for n b 2
n!y
by

ba ba i1

s1 ¼ a þ ; si ¼ a þ þ ðb  s1 Þ ð2 a i a n  1Þ:
log n log n n  1

We can see eðfDn gy

n¼1 Þ ¼ y and lim dðDn Þ ¼ 0.
n!y

Proof of Theorem 1.2. If eðfDn gy n¼1 Þ ¼ y, then there is nothing to prove.

So we assume that eðfDn gyn¼1 Þ < y. In this case lim dðDn Þ ¼ 0 holds. According
n!y
to the expression of the fourth derivative of F described in Lemma 2.2 there
exists a positive constant M with property that
 
q 4 F ðp; qÞ  
  
ð3:1Þ  aM
 qq 4 q¼r 

for any a a p < r < q a b. This inequality and Corollary 2.4 imply
 
X n 
2 2 
n jS  SðDn Þj ¼ 2pn  F ðsi1 ; si Þ
 i¼1 

Xn    !
1   1 q4F 
2  jðsi1 Þðsi  si1 Þ  þ 
3 4 
a 2pn 3!  4! qq 4 ðsi1 ; ri Þðsi  si1 Þ 
i¼1

p p
a max jjjn 2  ndðDn Þ 3 þ Mn 2  ndðDn Þ 4
3 ½a; b
12
p p
¼ max jjjfndðDn Þg 3 þ MfndðDn Þg 3  dðDn Þ;
3 ½a; b
12

where si1 < ri < si . Considering the limits of the above inequality we obtain

p 3
lim sup n 2 jS  SðDn Þj a eðfDn gy
n¼1 Þ max jjj;
n!y 3 ½a; b

which completes the proof of Theorem 1.2. r

 Approximate sums of the areas of surfaces of revolution 291

Corollary 3.3. If fDn gy n¼1 is a sequence of n-divisions of ½a; b
with property

y
that eðfDn gn¼1 Þ < y, then we have

lim SðDn Þ ¼ S:
n!y

4. Proof of Theorem 1.3

Theorem 1.2 holds for any sequence of n-divisions. On the other hand,
Theorem 1.3 holds for a sequence of optimal n-divisions and the estimate in it
is sharper than that in Theorem 1.2. In order to consider approximate sums of
optimal divisions we need the following lemma obtained by Gleason [2].

Lemma 4.1 (Gleason). Let FðtÞ be a nonnegative continuous function defined

on ½a; b
. For any positive integer n there exists a division of ½a; b
:
a ¼ s0 < s1 <    < sn1 < sn ¼ b
such that all of
ðsi  si1 Þ max FðtÞ ð1 a i a nÞ
½si1 ; si

are equal to each other. We denote by Jn the equal value. Then we obtain
ðb
lim nJn ¼ FðtÞ dt:
n!y a

Proof of Theorem 1.3. According to the mean value theorem there exists
p < r1 < q which satisfies

hðuÞ ¼ ðq  pÞf f 0 ðr1 Þ 2 þ 1g 1=2 :

This implies
jhðuÞj a ðb  aÞ maxfð f 0 Þ 2 þ 1g 1=2 ð0 a u a q  pÞ:
½ p; q

Using Lemma 2.1 we can see that there exist positive constants Mi ð1 a i a 4Þ
which satisfy
jhðiÞ ðuÞj a Mi maxfð f 0 Þ 2 þ 1g 1=2 ð0 a u a q  pÞ:
½ p; q

From these estimates and the expression of the fourth derivative of F in Lemma
2.2 there exits a positive constant M which satisfies
 
q 4 F ðp; qÞ  
   a M maxfð f 0 Þ 2 þ 1g 1=2

 qq 4 q¼r  ½ p; q

292 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

for any p < r < q. We define c by

1 1
cðxÞ ¼  f 00 ðxÞfð f 0 ðxÞÞ 2 þ 1g1 þ f ðxÞð f 00 ðxÞÞ 2 fð f 0 ðxÞÞ 2 þ 1g2 :
2 4

By the definitions of j and c we have

jðxÞ ¼ fð f 0 ðxÞÞ 2 þ 1g 1=2 cðxÞ:

The assumption that j > 0 implies c > 0. We set

m ¼ min c > 0:
½a; b

We multiply m a cðxÞ by fð f 0 ðxÞÞ 2 þ 1g 1=2 and obtain

mfð f 0 ðxÞÞ 2 þ 1g 1=2 a jðxÞ:

We apply Lemma 4.1 to the function j 1=3 and get an n-division DnG of ½a; b
which
satisfies all of

ðsi  si1 Þ max j 1=3 ð1 a i a nÞ

½si1 ; si

are equal to each other. We denote by Jn the equal value. Then we have
ðb
lim nJn ¼ jðtÞ 1=3 dt:
n!y a

Corollary 2.4 implies

n 2 jS  SðDnG Þj
   !
Xn 1   1 q4F 
2  jðsi1 Þðsi  si1 Þ  þ 
3 4 
a 2pn 3!  4! qq 4 ðsi ; ri Þðsi  si1 Þ 
i¼1

3 !
Xn
1 1
2 1=3 0 2 1=2 4
a 2pn max j ðsi  si1 Þ þ max fð f Þ þ 1g Mðsi  si1 Þ
i¼1
3! ½si1 ; si
4! ½si1 ; si

n

X 
1 M
a 2pn 2
Jn3 þ max jðsi  si1 Þ 4
i¼1
3! 4!m ½si1 ; si

p 2pM 1
¼ ðnJn Þ 3 þ  ðnJn Þ 3 ðb  aÞ;
3 4!m n
 Approximate sums of the areas of surfaces of revolution 293

where si1 < ri < si . Considering the limits of the above inequality we obtain

ð b 3
p
lim sup n 2 jS  SðDnG Þj a jðtÞ 1=3 dt :
n!y 3 a
G
Since jS  SðDa
n Þj a jS  SðDn Þj, we complete the proof of Theorem 1.3. r

5. Proof of Theorem 1.4

The assumption that f 00 < 0 implies that F ðp; qÞ > 0 for any subinterval
½p; q
H ½a; b
, because the orthogonal projection of the surface determined by
½p; q
to the corresponding frustum is area-decreasing. The condition f 00 < 0
implies j > 0, so we have already obtained the estimate of n 2 ðS  SðDn ÞÞ from
above. In this section we estimate n 2 ðS  SðDn ÞÞ from bellow.
In order to prove Theorem 1.4 it is su‰cient to prove the following lemmas.

Lemma 5.1. We assume f 00 < 0. For a sequence Dn of n-divisions of ½a; b
, we

have

ð b 3
p
lim inf n 2 ðS  SðDn ÞÞ b jðxÞ 1=3 dx :
n!y 3 a

Proof. First we prove that there exists a positive constant J independent of

p, q which satisfies
 
 1 
ð5:1Þ  3 4
F ðp; qÞ  6 jðxÞðq  pÞ  a Jðq  pÞ

for any interval ½ p; q
H ½a; b
and any x A ½p; q
. Corollary 2.4, (3.1), and the
mean value theorem imply
    
 1  1  1 q 4
F ðp; qÞ  

F ðp; qÞ  jðxÞðq  pÞ  ¼ ðq  pÞ ðjðpÞ  jðxÞÞ þ
3 3  ðq  pÞ
 6  6  4 qq 4  
q¼c

a Jðq  pÞ 4 :

Next we prove the following lemma, which implies Lemma 5.1. r

Lemma 5.2. For any e > 0 there exists a positive integer N such that for any
n b N and any n-division D of ½a; b
the following inequality holds.

ð b 3
p
n 2 ðS  SðDÞÞ b jðxÞ 1=3 dx  e:
3 a
294 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

Proof. We have
 
 F ð p; qÞ 1 
 
  jðxÞ  a Jðq  pÞ
ðq  pÞ 3 6 

by (5.1). Since the function x 7! x 1=3 is uniformly continuous on ½0; yÞ, for any
d > 0 there exists d1 > 0 such that jx  yj a d1 implies jx 1=3  y 1=3 j a d. We take
h > 0 with property that q  p a h implies Jðq  pÞ a d1 . If we take p < q which
satisfy q  p a h, then we have
 1=3 
F ðp; qÞ 1=3
1
 
  jðxÞ  a d;
 ðq  pÞ 6 

that is,

1=3 
 1 
 1=3 
ð5:2Þ F ð p; qÞ  jðxÞ ðq  pÞ a dðq  pÞ:
 6 

We take a positive integer r with property that ðb  aÞ=r a h. For any n-

division D of ½a; b
we can add at most r points to D such that the width
of each subinterval is less than or equal to h. We denote the new division
by

D0 : s0 ¼ a a s1 a    a st ¼ b;

where t a n þ r. According to the first mean value theorem for integration we can
take si0 in ½si1 ; si
satisfying
ð si
jðxÞ 1=3 dx ¼ jðsi0 Þ 1=3 ðsi  si1 Þ:
si1

By the estimate (5.2) we get

1
jðsi0 Þ 1=3 ðsi  si1 Þ a F ðsi1 ; si Þ 1=3 þ dðsi  si1 Þ:
6 1=3

We add the above inequalities for i ¼ 1; . . . ; t and get

ðb Xt
1 1
jðxÞ 1=3 dx ¼ jðsi0 Þ 1=3 ðsi  si1 Þ
6 1=3 a i¼1
6 1=3

X
t
a F ðsi1 ; si Þ 1=3 þ dðb  aÞ: ð *Þ
i¼1
 Approximate sums of the areas of surfaces of revolution 295

We apply the Hölder inequality to the first term of (*) and get
!1=3
1=3
X
t
1=3
X
t
ð2pÞ F ðsi1 ; si Þ a t 2=3 2pF ðsi1 ; si Þ
i¼1 i¼1

¼ t 2=3 ðS  SðD0 ÞÞ 1=3 :

From these we have

1=3 ð b
p
jðxÞ 1=3 dx a t 2=3 ðS  SðD0 ÞÞ 1=3 þ dð2pÞ 1=3 ðb  aÞ:
3 a

The inequality S  SðD0 Þ a S  SðDÞ, the estimate obtained above and t a n þ r

imply

1=3 ð b
p
jðxÞ 1=3 dx a ðn þ rÞ 2=3 ðS  SðDÞÞ 1=3 þ dð2pÞ 1=3 ðb  aÞ:
3 a

Using the result obtained above, we prove Lemma 5.1. Since the function
x 7! x 3 is continuous, for any e > 0 there exists r > 0 such that if

1=3 ð b
p
rb jðxÞ 1=3 dx  x;
3 a

then we have

ð b 3
e p 1=3
b jðxÞ dx  x3:
2 3 a

So we take d > 0 which satisfies dð2pÞ 1=3 ðb  aÞ a r. We can apply the result
obtained above and get a positive integer r with property that for any n-division
D of ½a; b


1=3 ð b
1=3 p
r b dð2pÞ ðb  aÞ b jðxÞ 1=3 dx  ðn þ rÞ 2=3 ðS  SðDÞÞ 1=3 ;
3 a

which implies

ð b 3
e p 1=3
b jðxÞ dx  ðn þ rÞ 2 ðS  SðDÞÞ:
2 3 a

We can substitute the optimal division Dan for D in the above inequality and get

ð b 3
2 a p 1=3 e
ðn þ rÞ ðS  SðD ÞÞ b jðxÞ dx  :
3 a 2
296 Yuriko Gorai, Hiroyuki Tasaki, and Mio Yamakawa

Since

lim supð2nr þ r 2 ÞðS  SðDaÞÞ ¼ 0;

n!y

we can choose a positive integer N with property that for n b N

e
0 a ð2nr þ r 2 ÞðS  SðDaÞÞ a
2

holds. Thus for n b N we have

e
n 2 ðS  SðDaÞÞ b ðn þ rÞ 2 ðS  SðDaÞÞ 
2

ð b 3
p 1=3
b jðxÞ dx  e:
3 a

Therefore for any n-division D of ½a; b
we have

ð b 3
2 p
2 a 1=3
n ðS  SðDÞÞ b n ðS  SðD ÞÞ b jðxÞ dx  e;
3 a

which completes the proof of Lemma 5.1. r

Proof of Theorem 1.4. Since the condition f 00 < 0 implies j > 0, by

Theorem 1.3 and Lemma 5.1 we have

ð b 3
p 1=3
jðxÞ dx a lim inf n 2 ðS  SðDa 2 a
n ÞÞ a lim sup n ðS  SðDn ÞÞ
3 a n!y n!y

ð b 3
p
a jðxÞ 1=3 dx ;
3 a

which completes the proof of Theorem 1.4. r

References

[ 1 ] Charles K. Chui, Concerning rates of convergence of Riemann sums, J. Approx. Theory 4

(1971), 279–287.
[ 2 ] Andrew M. Gleason, A curvature formula, Amer. J. Math. 101 (1979), 86–93.
[ 3 ] H. A. Schwarz, Sur une définition erronée de l’aire d’une surface courbe, Gesammelte
Mathematische Abhandlungen von H. A. Schwarz, Zweiter Band, Chelsea Publishing
Company, Bronx, New York, 1972, 309–311.
 Approximate sums of the areas of surfaces of revolution 297

[ 4 ] Hiroyuki Tasaki, Convergence rates of approximate sums of Riemann integrals, J. Approx.

Theory, 161 (2009), 477–490.

Ose Highschool, Hitachiohmiya

Ibaraki, 319-2401 Japan

Graduate School of Pure and Applied Science

University of Tsukuba, Tsukuba
Ibaraki, 305-8571 Japan
E-mail address: tasaki@math.tsukuba.ac.jp

Graduate School of Pure and Applied Science

University of Tsukuba, Tsukuba
Ibaraki, 305-8571 Japan
E-mail address: m.yamakawa@math.tsukuba.ac.jp