On natural approaches related to classical trigonometric inequalities

Author(s): Abd Raouf Chouikha1
14, Cour des Quesblais 35430 Saint-Pere, FRANCE.
Copyright © Abd Raouf Chouikha. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

In this paper, we establish sharp inequalities for trigonometric functions. We prove in particular for 0<x<π2 and any n5 0<Pn(x) < (sinx)2x3cotx<Pn1(x)+[(2π)2nk=3n1ak(2π)2n2k]x2n where Pn(x)=3=knakx2k+1 is a n-polynomial, with positive coefficients (k5), ak=22k2 (2k2)!(|B2k2|+(1)k+1(2k1)k), B2k are Bernoulli numbers. This improves a lot of lower bounds of sin(x)x and generalizes inequalities chains.
Moreover, bounds are obtained for other trigonometric inequalities as Huygens and Cusa inequalities as well as for the function
gn(x)=(sin(x)x)2(12(2xπ)2n+21(2xπ)2)+tan(x)x, n1.

Keywords: Trigonometric functions; Sinc function; Inequalities.

1. Introduction

Inequalities involving trigonometric functions are used in many applications in various fields of mathematics. The method to compare functions to their corresponding Taylor polynomials has been successfully applied to prove and approximate a lot of trigonometric inequalities [1].
A method called the natural approach, introduced by Mortici in [2], uses the idea of comparing functions to their corresponding Taylor polynomials. This method has been successfully applied to prove and approximate a wide category of trigonometric inequalities.

Let us consider the double inequality

(1) (cosx)13<sinxx<2+cosx3. The left-hand side is known as Adamovic-Mitrinovic inequality (see [1,3]), while the right-hand side is known as Cusa inequality. The latter one which was proved by Huygens was used in order to estimate the number π, [3].

In this paper we provide another natural approach by comparing functions with their corresponding Taylor polynomials. This approach is analog to that given by [2]. As a corollary, that permits us to extend and sharpen results related to trigonometric inequalities and give generalizations and refinements.

In particular, the following inequalities have recently been established

Statement 1. [4,Theorem 1 p.4] for 0<x<π/2 one has (2) x415+23x6189041x837800 < cosx(sinxx)3 < x415+23x6189041x837800+53x10831600.

Notice that this statement is finer that provided by Mortici [2] : x415< cosx(sinxx)3 < x415+23x61890.

Statement 2. [4,Theorem 3 p.7] for 0<x<π/2 one has 2sinxx+tanxx>3+(320x43140x6+32240x8119800x10)(cosx)1.

Statement 3. [5,Theorem] Neuman-Sandor for 0<x<π/2 one has 2sinxx+tanxx>2xsinx+xtanx>3.

Statement 4. Cheng and Paris proved (Theorem 3.4 (3.23) of [6]) 3+(320+1280x2+2333600x4) x3tanx<2sinxx+tanxx.

Statement 5. [4,Theorem 4 p.9] for 0<x<π/2 one has 2+1cosx(8x4454x6105+19x8472537x10133650)<(sin(x)x)2+tan(x)x<2+1cosx(8x4454x6105+19x84725).

In this paper, we aim to refine all the inequalities mentioned above.

2. Main results

In [7] one proved for  0<x<π2 (3) cosx+x3(1x263)sinx15 < (sinxx)3 < cosx+x3sinx15. In the sequel we provide lower bound of 1 which is finer than 3 and appears to be sharper than many known bounds .

To that end, we will consider a function f(x)=(sinx)2x3cotxx615+x8945 defined for 0<x<π/2 and we will provide good estimations.

2.1. Estimation for the function  f(x)

At first, consider the following technical lemma which has been proved by [[8] , Lemma 3.1], but the proof we give here is slightly different.

Lemma 1. For 0<x<π/2 consider the function f(x)=(sinx)2x3cotxx615+x8945 then f(x) can be expanded as power series f(x)=k5akx2k, ak=22k2|B2k2|(2k2)!+(1)k+122k1(2k)!=22k2(2k2)!(|B2k2|+(1)k+1(2k1)k), where B2k are the Bernoulli numbers. Moreover, the coefficients ak,k5 are all positive: a5=1255150, a6=215436575, a7=1038300667375,

Proof. The following series expansions can be found in [9]

cotx=1xk=122kB2k(2k)!x2k1,x(0,π) and sin2x=k=1(1)k+122k1(2k)!x2k,x(0,π2)

(sinx)2x3cotx=(115x61945x8+12835x10+8467775x12+20691216125x14+ 139638512875x16+10861488462349375x18+438628194896477400625x20+O(x22)). To prove the positivity of the coefficients ak,k even we will use the following inequality for Bernoulli numbers established by D’Aniello [10]: 2(2k)!π2k(22k1)<∣B2k∣<2(2k)!π2k(22k2). For any odd value of k, we have ak=22k2 (2k2)!(|B2k2|+(1)k+1(2k1)k)=22k2 (2k2)!(|B2k2|+1(2k1)k)>0. Consider the even case, then ak=22k2 (2k2)!(|B2k2|+1(2k1)k). By definition of Bernoulli numbers Sn(p)=k=0n1kp=1p+1k=0p(kp+1)Bknp+1k.

Then for k even ak=22k2 (2k2)!(|B2k2|1(2k1)k)>22k2 (2k2)!(2(2k2)!π2k2(22k22)1(2k1)k) and ak<22k2 (2k2)!(2(2k2)!π2k2(22k21)1(2k1)k). Therefore 22k22(2k2)!π2k2(22k21)(2k2)!+(1)k+122k1(2k)!<ak=22k2|B2k2|(2k2)!+(1)k+122k1(2k)! <22k22(2k2)!π2k2(22k22)(2k2)!+(1)k+122k1(2k)!. and k5(22k12π2k2(22k21)+(1)k+122k1(2k)!)x2k<f(x). Thus it implies the left hand of (4) since for any integer n5 we have cosxcosx+sinx (x315x5945)+sinx5knakx2k3 cosx+sinx (x315x5945)+sinx5kakx2k3=(sinxx)3.
Notice that (sinx)2x3cotx=(sinx)2x3(1xk=122kB2k(2k)!x2k1)= (sinx)2x2+k=122kB2k(2k)!x2k+2<(sinx)2x2+k=122k+1x2k+2π2k(22k2).

In the other hand we know that for k>1 (2k)!>4πk (2ke)2k e124k+1. It implies (2(2k2)!π2k2(22k21)1(2k1)k)>(24π(k1)(2k2e)2k2e148k23π2k2(22k21)1(2k1)k)> (24π(k1)(2k2πe)2k2(22k21)1(2k1)k)>(24π(k1)(2k2πe)2k222k21(2k1)k)> (24π(k1)(k1πe)2k21(2k1)k). Thanks to Maple we may easily verify that the last expression is non negative as soon as  k>6. This means that  ak are non negative. ◻

Our first result Theorem 1 motivated us to further refine the Adamovic-Mitrinovic inequality. It permits us to deduce the lower bound of (sinxx)3 which appears to be finer than the corresponding bounds in Statement 1.

Theorem 1.  

For 0<x<π2 for any n5 the following inequalities hold (4)sinxk=5nakx2k3  (sinxx)3cosxsinx (x315+x5945)sinxk=5n1akx2k3+sinx[(2π)2nk=5n1ak(2π)2n2k]x2n3,

where ak=22k2 (2k2)!(|B2k2|+(1)k+1(2k1)k),k5 and  B2k are the Bernoulli numbers.

Proof. By Lemma 1 the function f(x)=(sinx)2x3cotxx615+x8945 is positive. We also need the following Lemma which gives a upper and lower bound for the preceding function.

Lemma 2.  consider the real analytic functions f defined on the interval [a,b] : f(x)=k=0 ak(xa)k where akIR,ak0 for all kIN Then  f(x) may be bounded by Taylor’s approximation for any  n1 k=0nf(k)(a+)k!(xa)kf(x)k=0nf(k)(a+)k!(xa)k+1(ba)nRn(b)(xa)n, where  Rn(x)=f(x)k=0nf(k)(a+)k!(xa)k.

We may find an elegant proof of this Lemma in [15,Malesevic-Rasajski-Lutovac,Theorem 4] .

Moreover, if we suppose in addition f(n) is increasing in [a,b] we easily deduce k=0nf(k)(a+)k!(xa)k+f(n)(a+)n!(xa)nf(x) k=0nf(k)(a+)k!(xa)k+f(n)(a+)n!(xa)n+1(ba)nRn(b)(xa)n. The coefficients  f(n)(a+)n! and  1(ba)nRn(b) are the best possible constants. This result may also be deduced from [8,Theorem 3.2]

Applying the preceding to the function f(x)=(sinx)2x3cotx=k5akx2k we then derive for 0<x<π/2  and  n5 the following inequalities k5nakx2kf(x)k5n1akx2k+[(2π)2nk5n1ak(2π)2n2k]x2n hold. Therefore,

sinxk5nakx2k3f(x)sinxx3sinxk5n1akx2k3+ sinx[(2π)2nk5n1ak(2π)2n2k]x2n3. This means sinxk5nakx2k3(sinxx)3cosxx3sinx15+x5sinx945 sinxk5n1akx2k3+sinx[(2π)2nk5n1ak(2π)2n2k]x2n3. This proves the theorem. ◻

Some particular cases of Theorem 1 are given below. Let us introduce some examples of the inequalities obtained for n=5,6,7,8,.
Putting n=5 , we obtain the following

Proposition 1.  For 0<x<π/2 the following inequalities hold (5) cosx+x3(1x263+x4189)sinx15 < (sinxx)3<cosx+x3(1x263)sinx15+1024π10x7sinx, Moreover, for 0<x<π/2 the following inequalities hold (6)[1+cosx2]2<cosx+x3(1x263+x4189)sinx15 < (sinxx)3.

Taking n=6, one has for 0<x<π/2 the following inequality (7) cosx+x3(1x263+x4189+831185x6)sinx15 < (sinxx)3<
cosx+x3(1x263+x4189)sinx15+(4096π1242835π2)x9sinx.
Taking n=7, one has for 0<x<π/2 the following inequality (8) cosx+x315(1x263(1x238x4495206x696525))sinx<(sinx)3x3< cosx+x315(1x263(1x238x4495))sinx+(16384π14162835π4325π2)x11sinx.
Putting n=8, one has for 0<x<π/2 the following inequality (9) cosx+x315(1x263(1x238x4495206x696525139x8675675))sinx<(sinx)3x3< cosx+x315(1x263(1x238x4495206x696525))sinx+ (65536π16642825π6128467775π482491216125π2)x13sinx. Etc…

Remark 1. Notice that when the degree n of the polynomial increases, the functionsinx5knakx2k3approaches the upper bound(sinxx)3cosxx315+x5945  since the coefficients  ak>0. In other words, the precision increases with n and allows us to have a good estimate of the error.
Indeed, consider the difference cosx+x315(1x263(1x23+bx4))sinx(sinx)3x3 For b=8495 we have cosx+x315(1x263(1x238x4495))sinx(sinx)3x3 <206x1291216125+304x141915538625373x1678153975900<0
For a=13, b=8495 and c=20696525 we get cosx+x315(1x263(1x23+bx4+cx6))sinx(sinx)3x3 <139x14638512875+13723x16976924698750559483x181559171819205000<0
For a=13, b=8495, c=20696525, d=139675675 we get cosx+x315(1x263(1x238x4495206x696525139x8675675+dx10))sinx(sinx)3x3 <10861x16488462349375+567257x1838979295480125013763x20359808881355000<0

2.2. Another estimation for  f(x)

The following result provided new bounds for the function f(x)=(sinx)2x3cotx It has been proved by [[8] – Theorem 3.3, p.10]

Lemma 3.  For 0<x<π/2 and  n0  the following inequalities hold (sinx)k=1nbk+2x2k+1<(sinx)2x3cotx<(sinx)k=1n1bk+2x2k+1+ [(2π)2n+5k=1n1bk+2(2π)2n2k]x2n+3sinx where bk=2(2k1)(2k+1)(22k11)B2k+(1)k(2k+1)!. Moreover, all the coefficients bk are non negative for k2.

Let f(x)=(sinx)2x3cotx then following [8] one has the expansion f(x)=(115+19x21890+167x4113400+479x62494800+)x5sinx.
Here is an improvement of this Lemma which will be useful to refine the bounds of the function.

Lemma 4.  For 0<x<π/2 and  n0  the following inequalities hold (sinx)[k=1nbk+2x2k+1(x515x7945)(1+k=12(22k11)|B2k|x2k(2k)!)]< (sinx)2x3cotxx615+x8945<(sinx)[k=1n1bk+2x2k+1(x515x7945)(1+k=12(22k11)|B2k|x2k(2k)!)]+ [(2π)2n+5k=1n1(2π)2n2k]x2n+3sinx where bk=2(2k1)(2k+1)(22k11)B2k+(1)k(2k+1)!. Moreover, all the coefficients  bk are non negative for k2.

Indeed, since [[11], p.145] 1sinx=1x+k=12(22k11)|B2k|x2k1(2k)! this Lemma may be deduced from Lemma 3 in writing x415+x6945=(sinx)(x415+x6945)(1x+k=12(22k11)|B2k|x2k1(2k)!).
Now, we will proceed as above, we will use Taylor’s approximation for this function to provide bounds for  (sinxx)3cosx. By Lemma 3 we derive

Theorem 2.  For 0<x<π/2 the following inequalities hold (sinx)2k=1nbk+2x2k2<(sinx)3x3cosx<(sinx)2k=1n1bk+2x2k2+ [(2π)2n+5k=1n1bk+2(2π)2n2k]x2n(sinx)2 where bk=2(2k1)(2k+1)(22k11)B2k+(1)k(2k+1)!, B2k are the Bernoulli numbers.

2.3 Bounds of Adamovic-Mitrinovic inequalities

By Theorem 1 we are able to improve the left hand inequality 3. Indeed, one has the following cosx(sinxx)3 <sinx Pn(x)< x315(1x263)sin(x)=sinxP5(x) where Pn(x),n5 is the n-polynomial Pn(x)=5knakx2k2=5kn(22k2(2k2)!((1)k+1(k)(2k1)+B2k)x2k2.

Turn to statement 1. In [[4], Theorem 1] the authors proved for 0<x<π/2  x415+23x6189041x837800<cosx(sinxx)3 < x415+23x6189041x837800+53x10831600 The authors used the following frame k=22n(1)kA(k)x2k < cosx(sinxx)3 < k=22n+1(1)kA(k)x2k, where  A(k)=32k+332k396k288k274(2k+3)!.
By increasing the degree n of  Pn(x) in 1 , one can improve the precision so that the left bound obtained is better than the one provided by [4].
In this case, taking  n=7 is enough to prove statement 1 P7(x)=x315(1x263+x4189+8x631185)sinx.

recall that we have for x(0,π/2) xx36+x5120x75040 <sinx < xx36+x5120x75040+x9362880.

Then we obtain thanks to Maple x415+23x6189041x837800+53x10831600+x315(1x263+x4189+8x631185)sinx >x415+23x6189041x837800+53x10831600+x315(1x263+x4189+8x631185)(xx36+x5120x75040)= 47x12157172400+19x14261954000x16294698250=x12(705171x2+8x4)2357586000. It is easy to see that the last expression is non negative for 0<x<π/2.
Then for x(0,π/2) the following inequalities hold x415+23x6189041x837800+53x10831600 > x315(1x263+x4189+8x631185)sinx > cosx(sinxx)3. Thus, for the left hand our estimate appears to be finer than that provided by [4]. However, this is not the case for the right hand. indeed, we may verify that the expression is non positive x3sinx15+x5sinx945x7sinx2835(16384π14162835π432467775π2)x9sinx+ x41523x61890+41x837800<0, which means that the following holds for 0<x<π/2 x3sinx15+x5sinx945x7sinx2835(16384π14162835π432467775π2)x9sinx< x415+23x6189041x837800<cosx(sinxx)3.
Other examples We interest here in the right part on the inequality. By the Taylor approximation we get the bound [[11], p.145] (sinx]2<x2x43+2x645x8315+=12n+1(1)k+122k+1x2k(2k)!. Then Theorem 2 implies (sinx)3x3cosx<(sinx)2[k=1n1bk+2x2k2+[(2π)2n+5k=1n1bk+2(2π)2n2k]x2n]< (x2x43+2x645)[k=1n1bk+2x2k2+[(2π)2n+5k=1n1bk+2(2π)2n2k]x2n]. Putting n=4 one gets thanks to Maple (sinx)3x3cosx<x415+23x61890(1128350+2048π11152945π416728350π2)x8<x415+23x6189041x837800 since (1128350+2048π11152945π416728350π2)+41x837800=0.0032400<0.
Putting n=5 one gets (sinx)3x3cosx<x415+23x6189041x837800+(291134008192π13+608945π6+33414175π4+479623700π2)x10< x415+23x6189041x837800+53x10831600 since (291134008192π13+608945π6+33414175π4+479623700π2)53831600=0.0016403<0. Thus Theorem 2 provides a finer bound than the one given by [4].

Remark 2. By the same way we may improve another bound of [4]: for 0<x<π/2 cosx(sinxx)3 < x415+23x6189041x837800+53x1083160074677x1227243216000+989x1410897286400. We must then use in Theorem 1 the polynomial Pn(x) with a degree of order n=9 to get a better estimate. Using Maple consider x315sinx(1x263(1x238x4495206x696525))(x415+23x6189041x837800+53x1083160074677x1227243216000+989x1410897286400)> x315(xx36+x5120x75040)(1x263(1x238x4495206x696525)) (x415+23x6189041x837800+53x1083160074677x1227243216000+989x1410897286400)= x12544320014929x1470053984000+197x1612770257500103x18229864635000=x12(1351350+1567545x2113472x4+3296x6)7355668320000>0. That means for 0<x<π/2 the following hold cosx(sinxx)3 < x315sinx(1x263(1x238x4495206x696525))< x415+23x6189041x837800+53x1083160074677x1227243216000+989x1410897286400.

A conjecture More generally, all the examples studied above naturally suggest that we may expect that the following inequalities hold cosx(sinxx)3 < x6sinx15+x8sinx945sinx k=5nakx2k3 < k=22n+1(1)kA(k)x2k, where ak=22k2 (2k2)!(|B2k2|+(1)k+1(2k1)k),A(k)=32k+332k396k288k274(2k+3)!. We may also expect (following Theorem 2 that (sinx)3x3cosx<(sinx)2k=1n1bk+2x2k2+ [(2π)2n+5k=1n1bk+2(2π)2n2k]x2n(sinx)2<k=22n(1)kA(k)x2k hold.
Other refinements
We may prove the following frame which also improves the one of Mortici x6945+x81890x1019800<(sinx)3x3cosxx315sinx <x6945+x81890x1019800+2903x121135134000<0

The following are stronger x85670+113x1012474005293x12486486000<(sinx)3x3cosx x315sinx(1x263cosx)<x85670+113x101247400<0

For a1 we have (sinx)3x3cosxx3sinx15+x5sinx945ax7sinx2835 >(12835a2835)x8+(13311850+a17010)x10+(571243243000a340200)x12>0

For a=1 we have

(sinx)3x3cosx115x3sinx+1945x5sinx12835x7sinx >8x104677752x12337837531x141915538625+x1672499050043789x181039447879470000>0
In particular 8x104677752x123378375>0.

3. Huygens inequality

Notice that we have an equivalence between inequalities cosx<(sinxx)3  1<(sinxx)2  tanxx,0<x<π/2.

By using the arithmetic-geometric mean inequality, Baricz and Sandor have pointed out that this inequality implies 2sinxx+tanxx>3and(sinxx)2+tanxx>2 for 0<x<π/2.

The following inequality which is due to Huygens (10) frac2sinxx+tanxx>3,x(0,π)(ii), is a consequence of (1.1).
Mortici [2] showed 3+3x420cosx3x6140cosx<2sinxx+tanxx<3+3x420cosx which improves 9>. Neuman-Sandor [5] proved the following 2sinxx+tanxx>2xsinx+xtanx>3.

Cheng and Paris proved (Theorem 3.4 (3.23) of [6]) 3+(320+1280x2+2333600x4) x3tanx<2sinxx+tanxx.

The following result improves the one of Mortici [[2],p.]

Theorem 3.  For 0<x<π/2 the following inequalities holds 2sinxx+tanxx>sinxx (3+12x2+524x4+61120x6+277144x8)> 3+3x420cosx3x6140cosx>3.

Proof. Notice that for 0<x<π/2 we have the expansion [[11] p.140] 1cosx=1+k1E2k(2k)!x2k=1+12x2+524x4+61120x6+, where  E2k are Euler numbers. Then we have obviously the left hand of inequalities since 1cosx>1+12x2+524x4+61120x6+.
Write 2sinxx+tanxx=sinxx(2+1cosx)=sinxx(2+k1E2k(2k)!x2k)> sinxx(3+12x2+524x4+61120x6+277x8144).
 ◻

We need the lemma

Lemma 5.  For 0<x<π/2 the following trigonometric inequality holds 1cosx<(12xπ+4xπ2)(12xπ)1.

Indeed, the Euler numbers verify the following frame, [12] [AS, p.805] 4k+1π2n+1(1+32n1)<E2k(2k)!<4k+1π2n+1. We then deduce 1cosx=1+k1E2k(2k)!x2k<1+k14k+1π2n+1x2k= 1+2πk1(xπ)2k=1+2π[112π1]= [1+2π12π2π]=(12xπ+4xπ2)(12xπ)1.
On the other hand, recall that for 0<x<π/2 xx36+x5120x75040 <sinx. Therefore, thanks to Maple we can estimate the difference sinxx(3+x22+5x424+61x6120+277x8144)33x420cosx+3x6140cosx> (1x26+x4120x65040)(3+x22+5x424+61x6120+277x8144) 3+(3x420+3x6140)(12xπ+4xπ2)(12xπ)1= [(1x26+x4120x65040)(3+x22+5x424+61x6120+277x8144)(12xπ) 3(12xπ)+(3x420+3x6140)(12xπ+4xπ2)](12xπ)1= 277362880x15π277725760x144817151200x13π+4817302400x12+191363302400x11π191363604800x10 74212016x9π+74214032x8+(359360π+335π2)x7+359720x635x5π2(12xπ)1>0 for 0<x<π/2.
The right inequality of Theorem 3 is then proved.
Turn now to statement 2 2sinxx+tanxx>3+(320x43140x6+32240x8119800x10)(cosx)1.

Theorem 4.  For 0<x<π/2 the following inequalities hold 2sinxx+tanxx>(3+x22+5x424+61x6120+277x8144)> 3+(320x43140x6+32240x8119800x10)(cosx)1.

Proof. Write the difference h(x)=sinxx(3+x22+5x424+61x6120+277x8144)3(3x4203x6140+3x82240x1019800)1cosx. By Lemma 5 1cosx<(12xπ+4xπ2)(12xπ)1 implies h(x)>(1x26+x4120x75040)(3+x22+5x424+61x6120+277x8144) 3(3x4203x6140+3x82240x1019800)(12xπ+4xπ2)(12xπ)1= [(1x26+x4120x75040)(12xπ)(3+x22+5x424+61x6120+277x8144) 3(12xπ)(3x4203x6140+3x82240x1019800)(12xπ+4xπ2)](12xπ)1= (505219144576000π505214572288000π2)x175052118289152000x16+ (115679217728000π115679108864000π2)x15115679435456000x14+(423291555200π+42329777600π2)x13+ 423293110400x12+(1207221119958400π1097657907200π2)x111207221139916800x10+(185395040π+74211008π2)x9+ 1853910080x8+(359360π+481252π2)x7+359720x6+3x55π2= 0.0000006391x170.0000027623x16+0.00006146x150.00026565x14 0.0031484x13+0.013609x12+0.06995x110.30243x100.42497x9+ 1.8392x80.12404x7+0.49861x6+0.060792x5>0. ◻

That means the following inequalities hold and implying statement 2 2sinxx+tanxx>sinx(3+x22+524x4+61120x6+277144x8)x1> 3+(320x43140x6+32240x8119800x10)(cosx)1.

Theorem 5.  For 0<x<π/2 the following inequalities holds 2sinxx+tanxx>sinxx (3+12x2+524x4+61120x6+277144x8)>2xsinx+xtanx>3.

Proof. We need this lemma

Lemma 6. Consider the function g(x)=(sinx)2(3+1/2x2+524x4+61120x6+277144x8)2x2x2cosx defined for 0<x<π/2.
Then g(x) is non negative in this interval.

Indeed , since  cosx<1x22+x44x66!. then we have g(x)>(xx36+1120x515040x7)2(3+x22+524x4+61120x6+277144x8) 2x2x2(112x2+124x41720x6)> 374501604800x12+7612943200x10+7611680x8+215x6>0.
This implies that  g(x)xsinx>0. Or equivalently sinxx (3+12x2+524x4+61120x6+277144x8)(2xsinx+xtanx)>0. This completes the proof.

 ◻

Theorem 6.  For 0<x<π/2 the following inequalities holds 2sinxx+tanxx>sinxx (3+12x2+524x4+61120x6+277144x8)> 3+(320+1280x2+2333600x4)x3tanx.

Theorem 6 implies obviously statement 4.

Proof. For 0<x<π/2  recall that xx36+x5120x75040+x99!x1111! <sinx<xx36+x5120x75040+x99!, 1x22+x44!<cosx<1x22+x44!x66!. Therefore since  1cosx1+x22. we have the inequality thanks to Maple sinxx(3+x22+5x424+61x6120+277x8144)(3+(320+x2280+23x433600)x3tanx)= [sin2x2x(3+x22+5x424+61x6120+277x8144)3cosx(320+x2280+23x433600)x3sinx]1cosx> 12x(2x4x33+4x5158x7315+4x928358x11155925)(3+x22+5x424+61x6120+277x8144) 3+3x22(320+x2280+23x433600)x3(xx36+x5120x75040+x9362880)= 2775613300x18+10815493380472268800x166306127261954000x14+10484664494191264000x12 2428708319958400x10+9718760480x8+151360x6+x48>0. Theorem 5 is then proved. ◻

4. Wilker inequality

The following inequality (sin(x)x)2+tan(x)x>2,x(0,π2) due to Wilker [13] was intensively studied by many authors, e.g, [14-16]

Mortici [4] proved

2+(845x48105x6)(1cosx)<(sin(x)x)2+tan(x)x<2+(8x445cos(x)).

Theorem 7.  For 0<x<π/2 the following inequalities holds for  1mn and  pn (1k=m+122k+1π2k(22k2)x2kk=1m22kB2k(2k)!x2k+k=3nakx2k2)× (2+k=1p22k(22k2)B2k(2k)!x2k+k=p+1n22k+1(22k2)π2k(22k1)x2k)<(sin(x)x)2+tan(x)x< (1k=m+122k+1π2k(22k1)x2k+k=1m22kB2k(2k)!x2k+k=3akx2k2)× (2+k=1p22k(22k2)B2k(2k)!x2k+2(2xπ)2p+211(2xπ)2). where  B2k are the Bernoulli numbers.

Proof. Remark at first we may write obviously (sin(x)x)2(1+2xsin(2x))=(sin(x)x)2+tan(x)x.

Lemma 7.  For 0<x<π/2 the following inequalities holds for any integer m5 (sin(x)x)2<1k=m+122k+1π2k(22k1)x2k+k=1m22kB2k(2k)!x2k+k=3akx2k2, 1k=m+122k+1π2k(22k2)x2kk=1m22kB2k(2k)!x2k+k=3nakx2k2<(sin(x)x)2.
By Lemma 1 and Theorem 1 we may deduce that f(x)x2=(sin(x)x)2xcotxx415+x6945=k5akx2k2, where ak=22k2(2k2)!(|B2k2|+(1)k+1(2k1)k). Thus for any integer  n5 (sin(x)x)2=xcotx+x415x6945+k5akx2k2xcotx+x415x6945+k=5nakx2k2 since coefficients  ak>0..
Therefore since by [[9], p.] (11)cotx=1xx3x3452x5945.k=n22kB2k(2k)!x2k1,x(0,π), and by [D’Agnello] 2(2k)!π2k(22k1)<∣B2k∣<2(2k)!π2k(22k2). Then we deduce the inequalities k=n+122k+1π2k(22k2)x2k1<cotx1x+k=1n22kB2k(2k)!x2k1<k=n+122k+1π2k(22k1)x2k1. It follows for any integer  m5 (sin(x)x)2<1k=m+122k+1π2k(22k1)x2k+k=1m22kB2k(2k)!x2k+k=3akx2k2, 1k=m+122k+1π2k(22k2)x2kk=1m22kB2k(2k)!x2k+k=3nakx2k2<(sin(x)x)2.

Let us consider now expansions trigonometric functions with power series. We will use the Taylor expansions of sin(x), sin(x)=xx33!+x55!x77!+.+(1)k1x2k1(2k1)!+(1)ksinθx(2k+1)!x2k+1 where 0<θ<1. It is easy to remark that 1x23!+x45!x67!<sinxx<1x23!+x45!x67!+x89! for 0<x<π2. We then deduce bounds for  (sinxx)2 k=12p(1)k+122k+1(2k)!x2k2<(sinxx)2<k=12p+1(1)k+122k+1(2k)!x2k2.

On the other hand, we know that 1sinx=1x+x6+7x3360+=1x+k=1(22k2)B2k(2k)!x2k1. We then derive the following inequalities for any n1

 ◻

Lemma 8.  For 0<x<π/2 the following inequalities holds for any integer p1 2+k=1p22k(22k2)B2k(2k)!x2k+k=p+1n22k+1(22k2)π2k(22k1)x2k<1+2xsin2x< 2+k=1p22k(22k2)B2k(2k)!x2k+2(2xπ)2p+211(2xπ)2.

Indeed, 1+2xsin2x=2+k=122k(22k2)B2k(2k)!x2k>2+k=1n22k(22k2)B2k(2k)!x2k. Then  D’Agnolo inequalities (..) 2+k=1p22k(22k2)B2k(2k)!x2k+k=p+1n22k+1(22k2)π2k(22k1)x2k<1+2xsin2x< 2+k=1p22k(22k2)B2k(2k)!x2k+2k=p+122kπ2kx2k= 2+k=1p22k(22k2)B2k(2k)!x2k+2(2xπ)2p+211(2xπ)2. We may also prove the following frame 2(2xπ)2n+21(2xπ)22k=1n12k1(2xπ)2k<1+2xsin2x<21(2xπ)2. Finally, we get a lower bound for the product (sin(x)x)2(1+2xsin2x)>(1k=m+122k+1π2k(22k2)x2kk=1m22kB2k(2k)!x2k+k=3nakx2k2)× (2+k=1p22k(22k2)B2k(2k)!x2k+k=p+1n22k+1(22k2)π2k(22k1)x2k). The upper bound is (sin(x)x)2(1+2xsin2x)<(1k=m+122k+1π2k(22k1)x2k+k=1m22kB2k(2k)!x2k+k=3akx2k2)× (2+k=1p22k(22k2)B2k(2k)!x2k+2(2xπ)2p+211(2xπ)2). Theorem 7 is then proved.

Examples Let 0<x<π/2. By Theorem 7 we are able to precise the lower bound of the Huygens inequality in putting different values of  n,p.
– Taking n=3 and p=2 we find again a result of [2] 2+8x445<(sin(x)x)2+tan(x)x. – Taking n=4 and p=3 we find again 2+8x445+16x6315<(sin(x)x)2+tan(x)x. – Taking n=5 and p=3 we have 2+8x445+16x6315+104x84725<(sin(x)x)2+tan(x)x. – Taking n=6 and p=4 we have 2+8x445+16x6315+104x84725+592x1066825<(sin(x)x)2+tan(x)x. This permits to find again statement 5 a result of [[4], p.9]. indeed, since (see lemma 9 below) for 0<x<π/2 cosx>1x22!+x44!x66!+x88!x1010!+x1212!x1414! we then deduce (cosx)(2+8x445+16x6315+104x84725+592x1066825)>845x44105x6+194725x837133650x10. Taking n=7 and p=4 we obtain 2+8x445+16x6315+104x84725+592x1066825+152912x1242567525<(sin(x)x)2+tan(x)x. By the same way, using again the lower of  cosx we find a result of [4, p.10] (cosx)(2+8x445+16x6315+104x84725+592x1066825+152912x1242567525)> 845x44105x6+194725x837133650x10+28320638800x1235036810804000x14.

Etc…
In the sequel we will find upper and lower bounds of Huygens inequalities which appear to be finer than known previous. Consider at first

Lemma 9.  For 0<x<π/2 the following inequalities holds for any integer  p1 k=12p(1)k+122k+1(2k)!x2k2<(sinxx)2<k=12p+1(1)k+122k+1(2k)!x2k2, k=12p+1(1)k1(2k)!x2k<cosx<k=12p(1)k1(2k)!x2k.

Let us consider expansions trigonometric functions with power series. We will use the Taylor expansions of sinx, cosx sinx=xx33!+x55!x77!+.+(1)k1x2k1(2k1)!+(1)ksinθx(2k+1)!x2k+1, cosx=1x22!+x44!x66!+.+(1)kx2k(2k)!+(1)k+1cosθx(2k+2)!x2k+2 where 0<θ<1. It is easy to remark that 1x23!+x45!x67!<sinxx<1x23!+x45!x67!+x89! for 0<x<π2. We then deduce bounds for  (sinxx)2 cosx k=12p(1)k+122k+1(2k)!x2k2<(sinxx)2<k=12p+1(1)k+122k+1(2k)!x2k2. k=12p+1(1)k1(2k)!x2k<cosx<k=12p(1)k1(2k)!x2k.
We then derive the following which improves Theorem 7

Theorem 8.  For 0<x<π/2 the following inequalities holds for any  q1,1pn (k=12q(1)k+122k1(2k)!x2k2)×(2+k=1p22k(22k2)B2k(2k)!x2k+k=p+1n22k+1(22k2)π2k(22k1)x2k) <(sin(x)x)2+tan(x)x< (k=12q+1(1)k+122k1(2k)!x2k2)×(2+k=1p22k(22k2)B2k(2k)!x2k+2(2xπ)2p+211(2xπ)2), where  B2k are the Bernoulli numbers.

Corollary 1.  For 0<x<π/2 the following inequalities holds for any  n,p,1pn (sin(x)x)2×(2+k=1p22k(22k2)B2k(2k)!x2k+2(2xπ)2n+21(2xπ)22k=1n12k1(2xπ)2k)< (sin(x)x)2+tan(x)x< (sin(x)x)2×(2+k=1p22k(22k2)B2k(2k)!x2k+2(2xπ)2n+21(2xπ)2). where  B2k are the Bernoulli numbers.

Corollary 1 means that the following inequalities hold

(sin(x)x)2×(2+k=1p22k(22k2)B2k(2k)!x2k2k=1n12k1(2xπ)2k)< gn(x)=(sin(x)x)2(12(2xπ)2n+21(2xπ)2)+tan(x)x< (sin(x)x)2×(2+k=1p22k(22k2)B2k(2k)!x2k)<(sin(x)x)2+tan(x)x. The function  gn(x)  is growing as  n increasing. We have (sin(x)x)2(12(2xπ)21(2xπ)2)+tan(x)x<gn(x)<(sinxx)2+tanxx. Moreover, we may compute the limit when  x tends to π2 limxπ2gn(x)=2(5+4n)π2.

Examples  Let 0<x<π/2. – Taking n=3 and p=2 we find 2+8x445<(sinxx)2+tanxx(sinxx)2(2(2xπ)81(2xπ)2)< 2306x104677752x863+16x6315+8x445+2<16x6315+8x445+2. – Taking n=4 and p=3 we find 2+8x445+16x6315<(sin(x)x)2+tan(x)x(sin(x)x)2(2(2xπ)101(2xπ)2)< 61232x1230405375868x1066825+104x84725+16x6315+8x445+2<104x84725+16x6315+8x445+2. Taking n=5 and p=3 we obtain 2+8x445+16x6315+104x84725<(sin(x)x)2+tan(x)x(sin(x)x)2(2(2xπ)121(2xπ)2)< 1566172x141915538625480604x1291216125+592x1066825+104x84725+16x6315+8x445+2< 592x1066825+104x84725+16x6315+8x445+2. We then improve statement 5 ([4]) since by Lemma 9 (845x4+16315x6+1044725x8+59266825x10)cos(x)<845x44105x6+194725x8 (845x4+16315x6+1044725x8+59266825x10+15291242567525x12)cos(x)> 845x44105x6+194725x837133650x10+28320638800x12. Taking n=6 and p=3 we obtain 2+8x445+16x6315+104x84725+59266825x10<(sin(x)x)2+tan(x)x(sin(x)x)2(2(2xπ)141(2xπ)2)< 161934166x16488462349375123992x1458046625+152912x1242567525+592x1066825+104x84725+16x6315+8x445+2< 152912x1242567525+592x1066825+104x84725+16x6315+8x445+2 The last estimate improves ([4p.10]).

Etc…
Wu and Srivastava [15,Lemma 3] proved the following dual inequality (xsin(x))2+xtan(x)>2,0<x<π2 Mortici [4] proved

(xsin(x))2+xtan(x)>2+2x445.

(xsin(x))2+xtan(x)>2sin(x)x+tan(x)x

Mortici refined a result of Neuman and Sandor [5,Theorem 2.3], who showed

3xsinx+cosx>4,0<x<π2 establishing that 3xsinx+cosx>4+x410+x6210.

References

  1. Mitrinovic, D. S., & Adamovic, D. D. (1965). Sur une inegalite elementaire ou interviennent des fonc-tions trigonometriques, Univ u Beogradu. Publik. Elektr. Fakulteta. S. Matematika i Fizika, 149, 2334.
  2. Mortici, C. (2011). The natural approach of Wilker-Cusa-Huygens inequalities. Mathematical Inequalities & Applications, 14(3), 535-541.
  3. Mitrinovic, D. S., & Vasic, P. M. (1970). Analytic inequalities (Vol. 1). Berlin: Springer-verlag.
  4. Branko Malesevic , Tatjana Lutovac1, Marija Rasajski1 and Cristinel Mortici Extensions of the natural approach to refinements and generalizations of some trigonometric inequalities, Advances in Difference Equations,2018:90, (2018) https://doi.org/10.1186/s13662-018-1545-7
  5. Neuman, E., & Sandor, J. (2010). On some inequalities involving trigonometric and hyperbolic functions with emphasis on the Cusa-Huygens, Wilker, and Huygens inequalities. Mathematical Inequalities & Applications, 13(4), 715-723.
  6. Chen, C. P., & Paris, R. B. (2020). On the Wilker and Huygens-type inequalities. Journal of Mathematical Inequalities, 14(3), 685-705.
  7. Chouikha, A. R. (2021). New sharp inequalities related to classical trigonometric inequalities. Journal of Inequalities and Special Functions, 11(4), 27-35.
  8. Chen, C. P., & Cheung, W. S. (2012). Sharpness of Wilker and Huygens type inequalities. Journal of Inequalities and Applications, 2012, 1-11.
  9. Gradshteyn, I., Ryzhik, I.: Table of Integrals Series and Products, 8th edn. Academic Press, San Diego (2015) .
  10. D’aniello, C. (1994). On some inequalities for the Bernoulli numbers. Rendiconti del Circolo Matematico di Palermo Series 2, 43, 329-332. .
  11. Jeffrey, A., & Dai, H. H. (2008). Handbook of mathematical formulas and integrals. Elsevier.
  12. Abramowitz, M., & Stegun, I. (1972). Handbook of Mathematical Functions (Applied Mathematical Sciences. 55.) Washington.
  13. Wilker, JB: Problem E 3306. Am Math Mon. 96, 55 (1989). doi:10.2307/2323260.
  14. Wu, S. H., & Srivastava, H. M. (2007). A weighted and exponential generalization of Wilker’s inequality and its applications. Integral Transforms and Special Functions, 18(8), 529-535.

  15. Rasajski, M., Lutovac, T., & Malesevic, B. (2018). Sharpening and generalizations of Shafer-Fink and Wilker type inequalities: A new approach. Journal of Nonlinear Science and Applications, 11, 885-893.

  16. Wu, S. H., & Srivastava, H. M. (2008). A further refinement of Wilker’s inequality. Integral Transforms and Special Functions, 19(10), 757-765.