Skip to main content
Springer logoLink to Springer
. 2018 Apr 20;2018(1):94. doi: 10.1186/s13660-018-1687-x

On Frame’s inequalities

Ling Zhu 1,
PMCID: PMC5910469  PMID: 29706747

Abstract

In this paper, the errors of the two inequalities in Theorem 3.4.20 in the classic “Analytic Inequalities” by Mitrinovic are corrected, and the corresponding inequalities for circular functions and hyperbolic functions are rebuilt.

Keywords: Circular functions, Hyperbolic functions, Frame’s inequalities, Cusa–Huygens inequalities

Introduction

The classic “Analytic Inequalities” by Mitrinovic [1] has been hailed all over the world since it was published in 1970. The influence of this book on the various branches of mathematics cannot be overestimated and will last forever. As the author says in the introduction of his book, “The greater part of the results included have been checked, although this could not, of course, be done for all the results which appear in the book. We hope, however, that there are not many errors, but the very nature of this book is such that it seems impossible to expect it to be entirely free of them.” I find some errors in “Analytic Inequalities” and announce the specific contents.

The following conclusions are on p. 241 in [1].

Proposition 1

([1, Theorem 3.4.20])

For every t>0,

t3sint2+cost<1180t5, 1.1
t3sint2+cost(1+(1cost)29(3+2cost))<t72100. 1.2

It is not difficult to find that these two inequalities above on the common interval (0,π/2) are wrong. I read carefully the only one citation [2] by Frame in [1] for Theorem 3.4.20, which was published in the 1944 issue of “The American Mathematical Monthly” in the form of the report of the Mathematical Seminar. We judge that the object of Frame [2] is a right triangle, so the t must be the other two acute angles of a right triangle, that is, t(0,π/2). We can only find the related contents of (1.1) in [2] is the item “(7)”, but the one (1.2) at least did not appear in [2].

By using the analytic method, this paper has come to the corresponding conclusions of (1.1) and (1.2); specifically, these are, in the form of (1.1) and (1.2), the first of two inequalities holds for hyperbolic functions, while the second one must be reconstructed, and reversely for circular functions on the interval (0,π).

Theorem 1

Let x(0,π). Then

x3sinx2+cosx>1180x5 1.3

and

x3sinx2+cosx(1+(1cosx)29(3+2cosx))>x72100 1.4

hold, 1/180 and 1/2100 are the best constants in (1.3) and (1.4), respectively.

Theorem 2

Let x>0. Then

x3sinhx2+coshx<1180x5 1.5

and

x3sinhx2+coshx(1+(1coshx)29(3+2coshx))<(9.537610179105)x7 1.6

hold, where 1/180 is the best constant in (1.5).

Lemmas

Lemma 1

(Mitrinovic–Adamovic inequality [3])

The inequality

(sintt)3>cost 2.1

holds for all t(0,π/2), and the exponent 3 is best possible.

Lemma 2

(Lazarevic’s inequality [4])

Let t0. Then

(sinhtt)3>cosht 2.2

holds, and the exponent 3 is best possible.

Lemma 3

([5])

Let |x|<π, and B2n be the even-indexed Bernoulli numbers (see [6]), we have the following power series expansion:

1sinx=1x+n=122n2(2n)!|B2n|x2n1. 2.3

Lemma 4

([7])

Let an and bn (n=0,1,2,) be real numbers, and let the power series A(t)=n=0antn and B(t)=n=0bntn be convergent for |t|<R (R+). If bn>0 for n=0,1,2, , and if εn=an/bn is strictly increasing (or decreasing) for n=0,1,2, , then the function A(t)/B(t) is strictly increasing (or decreasing) on (0,R) (R+).

Lemma 5

([8])

Let B2n be the even-indexed Bernoulli numbers. Then the double inequality

π2(22n+21)(2n+2)(2n+1)(22n1)<|B2n||B2n+2|<π2(22n+11)(2n+2)(2n+1)(22n11) 2.4

holds for n=1,2, .

Lemma 6

Let B2n be the even-indexed Bernoulli numbers. Then the series

{δn}={n(22n+22)|B2n+2|(2n+2)(22n2)|B2n|} 2.5

is increasing for n1.

Proof

By Lemma 5 we have

δn1=(n1)(22n2)|B2n|(2n)(22n22)|B2n2|<(n1)(22n2)(2n)(22n22)(2n)(2n1)(22n21)π2(22n1),δn=n(22n+22)|B2n+2|(2n+2)(22n2)|B2n|>n(22n+22)(2n+2)(22n2)(2n+2)(2n+1)(22n11)π2(22n+11).

In order to prove cn1<cn for n2 it suffices to show

n(22n+22)(2n+2)(22n2)(2n+2)(2n+1)(22n11)π2(22n+11)>(n1)(22n2)(2n)(22n22)(2n)(2n1)(22n21)π2(22n1),

that is,

2n(2n+1)(22n8)(22n1)>(2n2)(2n1)(22n4)(22n2). 2.6

Let 2n=m. Then, for m4, we show

m(m+1)(2m8)(2m1)>(m2)(m1)(2m4)(2m2). 2.7

Let

um=m(m+1)(2m8)(2m1)(m2)(m1)(2m4)(2m2)um=2(2m1)22m(3m2+27m12)2m+32m16,vm=22m3m2+27m122m1.

Then

um>2(2m1)22m(3m2+27m12)2m=(2m1)2mvm 2.8

and

vm+12vm=317m2+2m3+m2(2m1)(2m+1)>0 2.9

for m4.

It follows from (2.9) and v4=80/7>0 that vm>0 for m4. Therefore, Lemma 6 follows from (2.6), (2.7) and (2.8). □

Lemma 7

The function

K(t)=t3sin3tt3sin2t

is increasing on (0,π). In particular, we have

  • (i)
    The double inequality
    1π38π3sin2t<(sintt)3<112sin2t 2.10
    holds for all t(0,π/2), the constants (π38)/π3 and 1/2 are best possible.
  • (ii)
    The inequality
    (sintt)3<1π38π3sin2t 2.11
    holds for all t(π/2,π), and the constant (π38)/π3 is best possible.

Proof

Let

K(t)=t3sin3tt3sin2t=(tsint)31t3sint:=A(t)B(t),0<t<π.

Then by (2.3) we obtain

1sin3t=1sint+12(1sint)=1t+n=122n2(2n)!|B2n|t2n1+1t3+n=2(22n2)(2n1)(n1)(2n)!|B2n|t2n3

and

A(t)=t3(1t+n=122n2(2n)!|B2n|t2n1)1+t3(1t3+n=2(22n2)(2n1)(n1)(2n)!|B2n|t2n3)=t2+n=1[(22n2)|B2n|(2n)!+(22n+22)(2n+1)n|B2n+2|(2n+2)!]t2n+2:=n=1ant2n+2,B(t)=t3(1t+n=122n2(2n)!|B2n|t2n1)=t2+n=1(22n2)|B2n|(2n)!t2n+2:=n=0bnt2n+2,

where

a0=1,an=(22n2)|B2n|(2n)!+(22n+22)(2n+1)n|B2n+2|(2n+2)!,b0=1,bn=(22n2)|B2n|(2n)!>0,n1.

Since

a0b0=1,anbn=1+n(22n+22)|B2n+2|(2n+2)(22n2)|B2n|=1+δn,n1,

we know a0/b0<a1/b1, and {an/bn}n1 is increasing by Lemma 6. So {an/bn}n0 is increasing, and K(t)=A(t)/B(t) is increasing on (0,π) by Lemma 4. In view of

K(0+)=12,K(π2)=π38π3,K(π)=+,

this completes the proof of Lemma 6. □

In order to prove (1.6), we need the following lemmas. We introduce a useful auxiliary function Hf,g. For a<b, let f and g be differentiable on (a,b) and g0 on (a,b). Then the function Hf,g is defined by

Hf,g:=fggf.

The function Hf,g has some good properties and plays an important role in the proof of a monotonicity criterion for the quotient of power series.

Lemma 8

([9])

Let C(t)=k=0cktk and D(t)=k=0dktk be two real power series converging on (r,r) (r+) and dk>0 for all k. Suppose that, for certain mN, the non-constant sequence {ak/bk} is increasing (resp. decreasing) for 0km and decreasing (resp. increasing) for km. Then the function C/D is strictly increasing (resp. decreasing) on (0,r) if and only if HC,D(r) (resp. ≤) 0. Moreover, if HC,D(r)< (resp. >) 0, then there exists t0(0,r) such that the function C/D is strictly increasing (resp. decreasing) on (0,t0) and strictly decreasing (resp. increasing) on (t0,r).

Lemma 9

Let

L(t)=sinh3tt3t3sinh2t,t>0,

Then the function L(t) has a minimum point t0=2.72078 , and

sinh3tt3t3sinh2tL(t0)=0.35803:=θ.

In particular, we see that the double inequality

(sinhtt)31+θsinh2t 2.12

holds for all t(0,+), the constant θ is best possible.

Proof

Let

L(t)=(sinhtt)31sinh2t=sinh3tt3t3sinh2t:=C(t)D(t),0<t<+.

Then by using the infinite series of sinhx and coshx we obtain

C(t)=sinh3tt3=14(sinh3t3sinht)t3=n=232n+134(2n+1)!t2n+1:=n=2cnt2n+1

and

D(t)=t3sinh2t=12(cosh2t1)t3=n=222n3(2n2)!t2n+1:=n=2dnt2n+1,

where

cn=32n+134(2n+1)!,dn=22n3(2n2)!>0,n2.

Setting

ζn:=cndn=32n+13(2n+1)(2n)(2n1)22n1,n2,

we have

ζ2=12>ζ3=1340>ζ4=205672,

and {ζn}n4 is increasing since

ζn+1ζn=34(10n229n12)32n+6n2+21n+1222nn(2n+3)(2n+1)(2n1)(n+1)>0

for n4. So

ζ2>ζ3>ζ4<ζ5<ζ6<.

We compute

HC,D(+)=limx+(CDDC)=+,

and we find that there exists t0(0,+) such that the function C/D is strictly decreasing on (0,t0) and strictly increasing on (t0,+) by Lemma 8. Let t1=2.72078,t2=2.72079. We calculate

L(t1)=3.9522×107<0,L(t2)=3.7644×107>0,

and see that there exists t0(t1,t2)=(2.72078,2.72079)(0,+) such that L(t0)=0. So

L(t0)=mint(0,+)L(t)=mint(2.72078,2.72079)L(t),

and

L(t0)L(t2)+(t0t2)L(t2)=0.35803.

Obviously, L(t)L(t0) implies (2.12). □

The proof of Theorem 1

The proof of the inequality (1.3)

Let

F(x)=x3sinx2+cosx1180x5,0<x<π.

Then

F(x)=136(cosx+2)2(x4(cosx+2)236(1cosx)2)=x2(cosx+2)+6(1cosx)36(cosx+2)2[x2(cosx+2)6(1cosx)].

In order to prove F(x)>0 holds for x(0,π), it suffices to show

x2(cosx+2)6(1cosx)<0,0<x<π. 3.1

Since

cosx=tan2x21tan2x2+1 3.2

we have

x2(cosx+2)6(1cosx)<01+3(tanx2)2<3(x2)2. 3.3

Let x/2=t. Then (3.3) is equivalent to

1+3tan2t<3t2,0<t<π2. 3.4

In fact, when letting

f(t)=3t213tan2t,0<t<π2, 3.5

then

f(0+)=1,f((π2))=12π210.21585>0. 3.6

By Lemma 1, we have

f(t)=6sin3t[(sintt)3cost]<0,0<t<π2, 3.7

which implies

f(t)mint(0,π/2)f(t)=f((π2))=12π21>0,0<t<π2. 3.8

So F(x)>0 holds for x(0,π), and

F(x)=x3sinx2+cosx1180x5>F(0)=0,0<x<π. 3.9

Since

limx0+x3sinx2+cosxx5=1180,

this completes the proof of the inequality (1.3). □

The proof of the inequality (1.4)

Let

G(x)=x3sinx2+cosx(1+(1cosx)29(3+2cosx))x72100,0<x<π.

Then G(0+)=0, and

G(x)=1300(2cosx+3)2[x6(2cosx+3)2200(1cosx)3]. 3.10

In order to prove G(x)>0 holds for x(0,π), it suffices to prove

x6<200(1cosx)3(2cosx+3)2,0<x<π. 3.11

Via (3.2) we have

2cosx+3=tan2x2+5tan2x2+1,1cosx=2tan2x2tan2x2+1,

and (3.11) is equivalent to

x6<(40tan3x2(secx2)(sec2x2+4))2. 3.12

So (3.12) holds for x(0,π) when proving

x3<40tan3x2(secx2)(sec2x2+4),0<x<π, 3.13

or

(x2)3<5tan3x2(secx2)(sec2x2+4),0<x<π. 3.14

Let x/2=t. Then t(0,π/2), and (3.14) is equivalent to

t3<5tan3t(sect)(sec2t+4)=5sin3t4cos2t+1, 3.15

or

(sintt)3>4cos2t+15=145sin2t. 3.16

In fact, by Lemma 7 we have

(sintt)3>1π38π3sin2t>145sin2t 3.17

for all t(0,π/2) due to 4/5>(π38)/π3=0.74199 .

Since

limx0+x3sinx2+cosx(1+(1cosx)29(3+2cosx))x7=12100,

this completes the proof of the inequality (1.4). □

So the proof of Theorem 1 is complete.

The proof of Theorem 2

The proof of the inequality (1.5)

Let

S(x)=x3sinhx2+coshx1180x5,0<x<+.

Then

S(x)=x4(coshx+2)236(coshx1)236(coshx+2)2=x2(coshx+2)+6(coshx1)36(cosx+2)2[x2(coshx+2)6(coshx1)].

In order to prove that S(x)<0 holds for x(0,+), it suffices to show

x2(coshx+2)6(coshx1)>0,0<x<+. 4.1

Since

coshx=tanh2x2+1tanh2x21 4.2

we have

x2(coshx+2)6(coshx1)>01+3(tanhx2)2>3(x2)2. 4.3

Let x/2=t. Then (4.3) is equivalent to

1+3tanh2t>3t2,0<t<+. 4.4

In fact, when letting

s(t)=3tanh2t13t2,0<t<+, 4.5

we have

s(0+)=1,s(+)=2. 4.6

By Lemma 2 we can obtain

s(t)=6sinh3t((sinhtt)3cosht)>0,0<t<+, 4.7

which implies

s(t)mint(0,+)s(t)=s(0+)=1>0,0<t<+. 4.8

So S(x)<0 holds for x(0,+), and

S(x)=x3sinhx2+coshx1180x5<S(0)=0,0<x<+. 4.9

Since

limx0+x3sinhx2+coshxx5=1180,

this completes the proof of the inequality (1.5). □

The proof of the inequality (1.6)

Let p=9.537610179105, and

H(x)=x3sinhx2+coshx(1+(1coshx)29(3+2coshx))+px7,0<x<+.

Then H(0+)=0, and

H(x)=21px6(2coshx+3)22(coshx1)33(2coshx+3)2.

We have

H(x)>0p<2(coshx1)321x6(2coshx+3)2H(x)>0x6<998.553028(coshx1)3(2coshx+3)2H(x)>0x6=998.553028(2tanh2x21tanh2x2)3(5tanh2x21tanh2x2)2=7988.424224cosh2x2tanh6x2(5tanh2x2)2H(x)>0x3<89.37798512coshx2tanh3x25tanh2x2=89.37798512sinh3x25cosh2xsinh2x2H(x)>0(sintt)3>5+4sinh2t11.17224814,

where t=x/2>0. In fact, by (2.12) in Lemma 9 we have

(sinhtt)3>1+0.35803sinh2t>5+4sinh2t11.17224814.

The last inequality holds for t(0,+) due to

11.17224814(1+0.35803sinh2t)(5+4sinh2t)=1.5642×109cosh2t+6.17224814>0.

Therefore H(x)>0, and H(x)>H(0+)=0 holds for x(0,+).

So the proof of Theorem 2 is complete. □

Remarks

Remark 1

The inequalities (1.3) and (1.4) are obviously better than the famous Cusa–Huygens inequality (see [1013]):

sinxx<2+cosx3,0<x<π2. 5.1

Remark 2

Mortici [14] strengthened (5.1) to

sinxx<2+cosx31180x4+13780x6,0<x<π2. 5.2

It is in Frame [2] that the following double inequality was also given:

2+cosxx2π23x2π2<sinxx<2+cosxx2103x210,0<x<π, 5.3

or

2+cosx3x(xsinx)3π2<sinxx<2+cosx3x(xsinx)30,0<x<π. 5.4

In order to compare the three inequalities (1.4), (5.2), and the right hand side of (5.4), we rewrite (1.4) as

sinxx<1x621001+(1cosx)29(3+2cosx)2+cosx3. 5.5

(i) We first compare two inequalities (5.5) and (5.2) on the same interval (0,π/2). We compute

(1+(1cosx)29(3+2cosx))(2+cosx31180x4+13780x6)(1x62100)2+cosx3=1170,100cosx+22cosx+3i(x),

where

i(x)=12,600cosx105x4cosx+59x6cosx1470x4+151x6+6300cos2x+6300.

Numerical results show that i(x)>0 for all x(0,0.0040) and i(x)<0 for all x(0.0040,π/2). That is, the upper estimate in (5.5) is smaller than the one in (5.2) on the interval (0,0.0040), meanwhile the upper estimate in (5.2) is smaller than the one in (5.5) on the interval (0.0040,π/2). So these two inequalities (1.4) and (5.2) are not included in each other.

(ii) Then we compare the two inequalities (5.5) and the right hand side of (5.4) on the same interval (0,π). Let us check the function

x621002+cosx3+(1cosx)29(3+2cosx)2+cosx3x(xsinx)30(1+(1cosx)29(3+2cosx))=118,900cosx+22cosx+3j(x),

where

j(x)=1400cosx70x2cosx+6x6cosx+980xsinx980x2+9x6+700cos2x+70xcosxsinx+700.

Numerical results show that j(x)>0 for all x(0,0.4878) and i(x)<0 for all x(0.4878,π). That is, the upper estimate in (5.5) is smaller than the one in the right hand side of (5.4) on the interval (0,0.4878), meanwhile the upper estimate in the right hand side of (5.4) is smaller than the one in (5.5) on the interval (0.4878,π). So these two inequalities (1.4) and the right hand side of (5.4) are not included in each other.

In a word, inequality (1.4) is not contained in the other improved Cusa–Huygens inequalities showed in [14] and [2] and is stronger than those ones near x=0.

Remark 3

Using the methods in [1517] and in [18], one can directly prove the inequalities (1.3) and (1.4), (1.5) and (1.6), respectively. A different approach based on the power series expansions, to proving, refinements and generalizations of inequalities of the similar type can be found in [19].

Conclusions

In the present study, we find that there are two wrong inequalities for circular functions in the famous monograph “Analytic Inequalities” by Mitrinovic, and we reestablish two inequalities on this topic and create two corresponding inequalities for hyperbolic functions. These new inequalities are the generalization of the famous Cusa–Huygens inequality, one of them is not contained in other improved Cusa–Huygens inequalities showed in [14] and [2] and is stronger than the ones near x=0.

Authors’ contributions

The author provided the questions and gave the proof for the main results. He read and approved the manuscript.

Funding

The author’s research is supported by the Natural Science Foundation of China grants No. 11471285 and the Natural Science Foundation of China grants No. 61772025.

Competing interests

The author declares that he has no competing interests.

Footnotes

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  • 1.Mitrinovic D.S. Analytic Inequalities. Berlin: Springer; 1970. [Google Scholar]
  • 2.Frame J.S. Some trigonometric, hyperbolic and elliptic approximations. Am. Math. Mon. 1954;61:623–626. doi: 10.2307/2307677. [DOI] [Google Scholar]
  • 3.Mitrinovic D.S., Adamovic D.D. Complement à l’article “Sur une inégalité élémentaire où interviennent des fonctions trigonométriques”. Publ. Elektroteh. Fak. Univ. Beogr., Ser. Mat. Fiz. 1966;166:31–32. [Google Scholar]
  • 4.Lazarevic I. Neke nejednakosti sa hiperbolickim funkcijama. Publ. Elektroteh. Fak. Univ. Beogr., Ser. Mat. Fiz. 1966;170:41–48. [Google Scholar]
  • 5.Li J.-L. An identity related to Jordan’s inequality. Int. J. Math. Math. Sci. 2006;6:76782. [Google Scholar]
  • 6.Scharlau W., Opolka H. From Fermat to Minkowski. New York: Springer; 1985. [Google Scholar]
  • 7.Biernacki M., Krzyz J. On the monotonicity of certain functionals in the theory of analytic functions. Ann. Univ. Mariae Curie-Skłodowska, Sect. A. 1955;2:134–145. [Google Scholar]
  • 8. Qi, F.: A double inequality for ratios of the Bernoulli numbers. http://www.researchgate.net/publication/264418049. 10.13140/RG.2.1.3461.2641
  • 9.Yang Z.H., Chu Y.M., Wang M.K. Monotonicity criterion for the quotient of power series with applications. J. Math. Anal. Appl. 2015;428(1):587–604. doi: 10.1016/j.jmaa.2015.03.043. [DOI] [Google Scholar]
  • 10.Campan F.T. The Story of Number π. Romania: Ed. Albatros; 1977. [Google Scholar]
  • 11. Iuskevici, A.P.: History of Mathematics in 16th and 16th Centuries, Moskva (1961)
  • 12.Cajori F. A History of Mathematics. 2. New York: MacMillan; 1929. [Google Scholar]
  • 13. Huygens, C.: Oeuvres Complètes. Publiées par la Societé Hollandaise des Science. Haga 1888–1940 (20 volumes)
  • 14.Mortici C. The natural approach of Wilker–Cusa–Huygens inequalities. Math. Inequal. Appl. 2011;14(3):535–541. [Google Scholar]
  • 15.Malesevic B., Makragic M. A method for proving some inequalities on mixed trigonometric polynomial functions. J. Math. Inequal. 2016;10(3):849–876. doi: 10.7153/jmi-10-69. [DOI] [Google Scholar]
  • 16.Lutovac T., Malesevic B., Mortici C. The natural algorithmic approach of mixed trigonometric-polynomial problems. J. Inequal. Appl. 2017;2017:116. doi: 10.1186/s13660-017-1392-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Malesevic, B., Lutovac, T., Banjac, B.: A proof of an open problem of Yusuke Nishizawa for a power-exponential function. J. Math. Inequal. (2018, in press)
  • 18.Makragic M. A method for proving some inequalities on mixed hyperbolic-trigonometric polynomial functions. J. Math. Inequal. 2016;11(3):817–829. [Google Scholar]
  • 19.Malesevic B., Lutovac T., Rasajski M., Mortici C. Extensions of the natural approach to refinements and generalizations of some trigonometric inequalities. Adv. Differ. Equ. 2018;2018:90. doi: 10.1186/s13662-018-1545-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Journal of Inequalities and Applications are provided here courtesy of Springer

RESOURCES