The Best Possible Constants Approach for

arXiv:2404.05733v1 [math.GM] 21 Mar 2024

Wilker-Cusa-Huygens Inequalities
via Stratification

April 10, 2024

Bojan Banjac a, Branko Malešević b, Miloš Mićović b, ∗,

Bojana Mihailović b, Milica Savatović b

a) Faculty of Technical Sciences, University of Novi Sad,

Trg Dositeja Obradovića 16, 21000 Novi Sad, Serbia
b) School of Electrical Engineering, University of Belgrade,
Bulevar kralja Aleksandra 73, 11000 Belgrade, Serbia

Abstract. In this paper, we generalize Cristinel Mortici’s results on Wilker-Cusa-

Huygens inequalities using stratified families of functions and SimTheP – a system for
automated proving of MTP inequalities.
Keywords: Wilker-Cusa-Huygens inequalities, stratified families of functions, a minimax ap-
proximant, automated proving of MTP inequalities, SimTheP

MSC 2020: Primary 68V15, 41A44; Secondary 26D05

1 Introduction
The basis of this research is well-known C. Mortici’s paper [1] in which the following
theorems have been proved:

Theorem 1 For every 0 < x < π/2, we have:

 3
x4 sin x x4 23x6
− < cos x − <− + .
15 x 15 1890

Theorem 2 For every 0 < x < π/2, we have:

1 4 sin x cos x + 2 1 4 1 6
− x < − <− x + x .
180 x 3 180 3780
∗ correspoding author

1
Theorem 3 For every 0 < x < π/2, we have:
 4 
3x 3x6 1 sin x tan x 3x4
3+ − <2 + <3+ .
20 140 cos x x x 20 cos x
Theorem 4 For every 0 < x < π/2, we have:
 4   2
8x 4x6 1 sin x tan x 8x4
2+ − < + <2+ .
45 105 cos x x x 45 cos x
Theorem 5 For every 0 < x < π/2, we have:
 x 2 x 2 4
+ > 2+ x .
sin x tan x 45
Theorem 6 For every 0 < x < π/2, we have:
x 1 4 1 6
3 + cos x > 4 + x + x .
sin x 10 210
Using a method based on the stratified families of functions described in the paper
[2], we show that it is possible to enhance inequalities presented in Theorems 1-6
up to the level of the best possible constants. Also, we show that from some of the
introduced families of functions, it is possible to single out corresponding minimax
approximations.

2 Preliminiares
In this section, we present assertions from the paper [2] and a method from the paper
[3] on the basis of which our main results will be obtained.

2.1 Stratified families of functions

Let
ϕp (x) : (a, b) −→ R
be a family of functions with a variable x ∈ (a, b) and a parameter p ∈ R+ .
A family of functions ϕp (x) is increasingly stratified if
(∀p1 , p2 ∈ R+ ) p1 < p2 ⇐⇒ ϕp1 (x) < ϕp2 (x)
holds for any x ∈ (a, b) and, conversely, it is decreasingly stratified if
(∀p1 , p2 ∈ R+ ) p1 < p2 ⇐⇒ ϕp1 (x) > ϕp2 (x)
holds for any x ∈ (a, b).
In this paper, we call sup |ϕp (x)| an error and denote it by d(p) .
x∈(a,b)

In [2], the conditions for the existence of the unique value p0 of the parameter p ∈ R+ ,
for which an infimum of an error is attained, are considered. Such infimum is denoted
by
d0 = inf sup |ϕp (x)|.
p∈R+ x∈(a,b)

For such a value p0 , the function ϕp0 (x) is called the minimax approximant on (a, b).

2
Theorem 7 (see Thm 1 from [2]) Let ϕp (x) be a family of functions that are contin-
uous with respect to x ∈ (a, b) for each p ∈ R+ and increasingly stratified for p ∈ R+ ,
and let c, d be in R+ such that c < d. If :
(a) ϕc (x) < 0 and ϕd (x) > 0 for all x ∈ (a, b), and at the endpoints ϕc (a+) =
ϕc (b−) = ϕd (a+) = 0 and ϕd (b−) ∈ R+ hold;
(b) the functions ϕp (x) are continuous with respect to p ∈ (c, d) for each x ∈ (a, b)
and ϕp (b−) is continuous with respect to p ∈ (c, d) too;
(c) for all p ∈ (c, d), there exists a right neighbourhood of the point a in which
ϕp (x) < 0 holds;
(d) for all p ∈ (c, d) the function ϕp (x) has exactly one extremum t(p) on (a, b), which
is minimum;
then there exists exactly one solution p0 , for p ∈ R+ , of the following equation

|ϕp (t(p) )| = ϕp (b−)

and for d0 = |ϕp0 (t(p0 ) )| = ϕp0 (b−) we have

d0 = inf sup |ϕp (x)| .

p ∈ R+ x ∈ (a, b)

The analogous theorem can be stated for decreasingly stratified functions.

Theorem 7’ (see Thm 1’ from [2]) Let ϕp (x) be a family of functions that are con-
tinuous with respect to x ∈ (a, b) for each p ∈ R+ and decreasingly stratified for p ∈ R+ ,
and let c, d be in R+ such that c < d. If :
(a) ϕc (x) > 0 and ϕd (x) < 0 for all x ∈ (a, b), and at the endpoints ϕc (a+) =
ϕd (b−) = ϕd (a+) = 0 and ϕc (b−) ∈ R+ hold;
(b) the functions ϕp (x) are continuous with respect to p ∈ (c, d) for each x ∈ (a, b)
and ϕp (b−) is continuous with respect to p ∈ (c, d) too;
(c) for all p ∈ (c, d), there exists a right neighbourhood of the point a in which
ϕp (x) < 0 holds;
(d) for all p ∈ (c, d) the function ϕp (x) has exactly one extremum t(p) on (a, b), which
is minimum;
then there exists exactly one solution p0 , for p ∈ R+ , of the following equation

|ϕp (t(p) )| = ϕp (b−)

and for d0 = |ϕp0 (t(p0 ) )| = ϕp0 (b−) we have

d0 = inf sup |ϕp (x)| .

p ∈ R+ x ∈ (a, b)

Theorem 8 (Nike theorem, see Thm 3 from [2] and Thm 2.1. from [4])
Let f : (0, c) −→ R be m times differentiable function (for some m ≥ 2, m ∈ N)
satisfying the following conditions:
(a) f (m) (x) > 0 for x ∈ (0, c);
(b) there is a right neighbourhood of zero in which the following inequalities are true:

f < 0, f ′ < 0, . . . , f (m−1) < 0;

3
 (c) there is a left neighbourhood of c in which the following inequalities are true:

f > 0, f ′ > 0, . . . , f (m−1) > 0.

Then the function f has exactly one zero x0 ∈ (0, c), and f (x) < 0 for x ∈ (0, x0 ) and
f (x) > 0 for x ∈ (x0 , c). Also, the function f has exactly one local minimum on the

interval (0, c). More precisely, there is exactly one point t ∈ (0, c) in fact t ∈ (0, x0 )
such that f (t) < 0 is the smallest value of the function f on the interval (0, c) and
particularly on (0, x0 ).

In the case when it is not possible to apply the Nike theorem, the following theorem is
applied, which gives sufficient conditions that the function on the interval has exactly
one zero and exactly one minimum (see section 3).

Theorem 9 (The Second Nike theorem, see Thm 4 from
[2]) Let f : (0, c) −→ R
be m times differentiable function for some m ≥ 2, m ∈ N satisfying the following
conditions:
(a) f (m) has exactly one zero xm on (0, c) such that f (m) > 0 on (0, xm ) and f (m) < 0
on (xm , c);
(b) there is a right neighbourhood of zero in which the following inequalities are true:

f < 0, f ′ < 0, ..., f (m−1) < 0;

(c) there is a left neighbourhood of c in which the following inequalities are true:

f > 0, f ′ > 0, ..., f (m−1) > 0.

Then the function f has exactly one zero x0 ∈ (0, c) and f (x) < 0 for x ∈ (0, x0 ) and
f (x) > 0 for x ∈ (x0 , c). The function f has exactly one minimum

on the interval (0, c),
i.e. there is exactly one point t ∈ (0, c) in fact t ∈ (0, x0 ) such that f (t) < 0 is the
smallest value of the function f on the interval (0, c) and particularly on (0, x0 ).

Remark 1 Let us emphasize that the previous two forms of the Nike theorem ensure
the existence of a minimax approximant. Also, these two theorems claim that the local
minimum at t is the only extremum of the function f on (0, c), which is shown in their
proofs (see [2]).

The topic of stratification appears in recently published papers [5]-[8].

2.2 A method for proving MTP inequalities

MTP function is determined by:
n
X
f (x) = αi xpi cosqi x sinri x ,
i=1

where x ∈ S ⊆ R (S is an open or closed interval), αi ∈ R\{0}, pi , qi , ri ∈ N0 and n ∈ N.

The corresponding inequality
f (x) > 0 ,
where x ∈ S, is called an MTP inequality.

4
MTP functions were originally considered through MTP systems of such functions
involving multiple variables, see [9]. In the article [3], the previous definitions of MTP
function and MTP inequality were introduced. Moreover, in that article, a method for
proving MTP inequalities on the base interval (0, π/2) was presented. The method is
based on Maclaurin approximations of the sine and cosine functions. Let us emphasize
that in [10], a method for proving MTP inequalities on the base interval, based on
universal trigonometric substitution and Maclaurin approximations, was presented, see
also [11]. It is noteworthy that both methods have been applied earlier in numerous
papers and monographs, see, for example, [1], [12], [13]. The topic of MTP functions
and stratified families of functions has been the subject of recently defended doctoral
dissertations [14], [15], [16].

In further consideration, let f (x) be an MTP function with rational coefficients. The
method for proving MTP inequalities from the paper [3] has been computationally im-
plemented through the doctoral dissertation [15]. For methods for proving inequalities
by computer, see also the papers [17] and [18]. The computer implementation from
[15], called SimTheP, proves MTP inequalities on the interval S ⊆ [0, π/2], providing
users with the proof in four stages. Therefore, we will outline the method through a
brief description of each of these stages based on the papers [3], [8], [19]:

I Recognition of possible case

In the first phase, the hypothesis f (x) > 0 on S is tested based on the values of the
function at the boundary points of the interval.

II Transformation of angles

In this phase, each addend of the MTP function f (x) undergoes the substitution of
the expression cosm x · sinn x (m, n ∈ N0 ) into a sum of sine and cosine functions of
multiple angles according to Table 1. These substitutions were proven in the paper
[3].

cosn x sinm x
n = qi m = ri Substitution
n m
2 −1 P
2 +P k (−1) m2 ( j )(k−j
+k+j n m
)cos((n+m−2k)x)
2n+m−1
k=0 j=0
even even n
+m
2P
m+ n +j n
2 (−1) 2 ( )( nj
m
+ m −j )
+ 2n+m
2 2

j=0
n m 1
P
2 + 2 −2 P
k (−1)
m +k+j n
2 ( )(k−j
j
m
)cos((n+m−2k)x)
odd even 2n+m−1
k=0 j=0
n m 1
P
2 + 2 −2 P
k (−1)
m +k+j− 1
2 2 (nj)(k−j
m
)sin((n+m−2k)x)
even odd 2n+m−1
k=0 j=0
n m
P
2 + 2 −1 Pk (−1) m
2
+k+j− 1
2 (nj)(k−j
m
)sin((n+m−2k)x)
odd odd 2n+m−1
k=0 j=0

Table 1: Substitutions of terms xpi cosqi x sinri x, see Table 1 from [19]

5
The initial MTP function is transformed into the equivalent form:
 
n mi  
X X (q ,r )
αi xpi 
 
(1) f (x) = θk trigk i i (qi − ri − 2k) x  ,
i=1 k=0
| {z }
(=t)

where 
(q ,r ) cos : qi -odd, ri -even or qi -even, ri -even
trigk i i =
sin : qi -odd, ri -odd or qi -even, ri -odd
and lq + r m
i i
mi = mi (qi , ri ) = − 1.
2
III Determination of downward rational polynomial approximation
During this phase, it is necessary to determine a downward polynomial approximation
with rational coefficients of the MTP function (1).
Let us specify the general concept of downward/upward polynomial approximation of
a function. Let φ(x) : A −→ R be any function defined over A ⊆ R. The downward
polynomial approximation of the function φ(x) over B ⊆ A is a polynomial P (x) such
that
(∀x ∈ B) φ(x) ≥ P (x) ,
denoted by P (x). Similarly, the upward polynomial approximation of the function
φ(x) over B ⊆ A is a polynomial P (x) such that

(∀x ∈ B) φ(x) ≤ P (x) ,

denoted by P (x).
In the following Lemma from the paper [8], we provide some upward and downward
polynomial approximations of the sine and cosine functions. These assertions were
proven in the paper [3].

Lemma 1 (see Lemma 1 from [8]) It holds:

(a) For the polynomial
(n−1)/2
X (−1)i t2i+1
Tn (t) = ,
i=0
(2i + 1)!
where n = 4k + 1, k ∈ N0 , it holds:
 h p i
∀t ∈ 0, (n + 3)(n + 4) T n (t) ≥ T n+4 (t) ≥ sin t ,
 h p i
∀t ∈ − (n + 3)(n + 4), 0 T n (t) ≤ T n+4 (t) ≤ sin t .
p
For t = 0, the inequalities turn into equalities. For t = ± (n + 3)(n + 4), the equali-
ties T n (t) = T n+4 (t) and T n (t) = T n+4 (t) hold, respectively.
(b) For the polynomial
(n−1)/2
X (−1)i t2i+1
Tn (t) = ,
i=0
(2i + 1)!

6
where n = 4k + 3, k ∈ N0 , it holds:
 h p i
∀t ∈ 0, (n + 3)(n + 4) T n (t) ≤ T n+4 (t) ≤ sin t ,
 h p i
∀t ∈ − (n + 3)(n + 4), 0 T n (t) ≥ T n+4 (t) ≥ sin t .
p
For t = 0, the inequalities turn into equalities. For t = ± (n + 3)(n + 4), the equali-
ties T n (t) = T n+4 (t) and T n (t) = T n+4 (t) hold, respectively.
(c) For the polynomial
n/2
X (−1)i t2i
Tn (t) = ,
i=0
(2i)!
where n = 4k, k ∈ N0 , it holds:
 h p p i
∀t ∈ − (n + 3)(n + 4), (n + 3)(n + 4) T n (t) ≥ T n+4 (t) ≥ cos t .
p
For t = 0, the inequalities turn into equalities. For t = ± (n + 3)(n + 4), the equality
T n (t) = T n+4 (t) holds.
(d) For the polynomial
n/2
X (−1)i t2i
Tn (t) = ,
i=0
(2i)!
where n = 4k + 2, k ∈ N0 , it holds:
 h p p i
∀t ∈ − (n + 3)(n + 4), (n + 3)(n + 4) T n (t) ≤ T n+4 (t) ≤ cos t .
p
For t = 0, the inequalities turn into equalities. For t = ± (n + 3)(n + 4), the equality
T n (t) = T n+4 (t) holds.

Let Tnφ,a (x) denote the Taylor expansion of order n of some analytic function φ in the
neighbourhood of some point a.
With the aim of obtaining a downward polynomial approximation P (x) of the MTP
function f (x), we approximate each addend of the function (1) by a Maclaurin poly-
nomial using the following estimates:
 cos,0

 αi θk > 0 : cos t > T4ℓ 1 +2
(x),

 αi θk < 0 : cos t < T cos,0 (x),

4ℓ2
(∗) sin,0
 αi θk > 0 : sin t > T4ℓ (x),

 3 +3

 sin,0
αi θk < 0 : sin t < T4ℓ4 +1 (x);

where t = (qi − ri − 2k) x and ℓ1,2,3,4 ∈ N0 .

By applying (∗), we determine a polynomial P (x) such that

f (x) > P (x)

for x ∈ S. If there exists a polynomial P (x) with rational coefficients such that

P (x) > 0

7
for x ∈ S, then
f (x) > 0
for x ∈ S.
If the coefficients of the MTP function f (x) are not rational numbers but computable
real numbers, then we could determine a downward polynomial approximation with
rational numbers, see the paper [20].
IV The final part
For real polynomials defined over a segment with endpoints where the polynomial does
not have zero, Sturm’s theorem provides the number of roots on such a segment, see,
for example, Theorem 4.1 [21] or originally [22]. It is particularly noteworthy that
for polynomial functions with rational coefficients defined over a segment with rational
endpoints, according to Theorem 4.2 [21], the problem of determining the number of
roots over that segment is an algorithmically decidable problem. For such polynomial
functions, if we obtain a proof of positivity using Sturm’s theorem, we can consider it
as an effective proof by finite procedures that can be manually verified.
In the third part, P (x) is determined as a polynomial with rational coefficients. If
S is not a segment with rational endpoints or the polynomial P (x) has a root at
the boundary points of the segment S, we consider the polynomial over an extended
segment with rational endpoints, see [23]. It is always possible to choose such a segment
with rational endpoints that the polynomial P (x) does not have a root at the boundary
points of that segment. If the number of roots does not increase over such an extended
segment, and we know whether the polynomial has a root at the boundary points of the
segment S, then we also have an effective proof of the polynomial inequality P (x) > 0
over S by applying Sturm’s theorem.
Let us emphasize that proofs based on the Sturm algorithm are absolutely theoretically
rigorous, as pointed out in the paper [24], see also [25]-[27].
In Appendix sections A1 and A4 - A7, we prove MTP inequalities f (x) > 0 over the
base interval (0, π/2), while in section A2, we prove MTP inequality over the interval
(0, 1] and in A3 over the interval [1, π/2].

3 Main results
According to the paper [2], the following statements, which are improvements of The-
orems 1–6 from the paper [1], are proved. Note that the automatic prover SimTheP
was utilized for proving the MTP inequalities. The results obtained by this prover are
provided in the Appendix.

Improvement of Theorem 1

Lemma 2 The family of functions

 3  
sin x 1 4
ϕp (x) = − cos x + − x + p x6 for x ∈ (0, π/2)
x 15

is increasingly stratified with respect to parameter p ∈ R+ .

8
Let us introduce the function g(x) so that the equivalence

ϕp (x) = 0 ⇐⇒ p = g(x), x ∈ (0, π/2), p ∈ R+

holds. Then
x7 + 15 x3 cos x − 15 sin3 x
g(x) = , x ∈ (0, π/2) .
15x9
Note that

ϕp (x) = p − g(x) x6 .

Lemma 3 The function g(x) is strictly decreasing for x ∈ (0, π/2).

Proof. Let us notice that the derivative g ′ is

45 x cos3 x−135 cos2 x sin x− 90 x3 +45 x cos x−15 x4 sin x+135 sin x−2 x7
g ′ (x) = .
15 x10
It holds
g ′ (x) < 0, x ∈ (0, π/2) ⇐⇒ f (x) > 0, x ∈ (0, π/2) ,
where


f (x) = 2 x7 − 45 x cos3 x + 135 cos2 x sin x + 90 x3 +45 x cos x + 15 x4 sin x − 135 sin x.

According to [3], there exists proof that the mixed trigonometric polynomial function
f is positive for x ∈ (0, π/2). The proof is given in Appendix A1. 

Statement 1 Let
4π 7 − 7680 23
A= = 0.0098430 . . . and B= = 0.012169 . . . .
15π 9 1890
Then, it holds:
(i) If p ∈ (0, A], then
 3
1 4 1 sin x
x ∈ (0, π/2) =⇒ − x + px6 ≤ − x4 + Ax6 < cos x − .
15 15 x

(p)
(ii) If p ∈ (A, B), then ϕp (x) has exactly one zero x0on (0, π/2). Also,
 3
(p) 1 sin x
x ∈ (0, x0 ) =⇒ − x4 + p x6 < cos x −
15 x
and  3
(p) sin x 1 4
x ∈ (x0 , π/2) =⇒ cos x − <− x + p x6
x 15
hold.
(iii) If p ∈ [B, ∞), then
 3
sin x 1 4 1
x ∈ (0, π/2) =⇒ cos x − <− x + Bx6 ≤ − x4 + px6 .
x 15 15

(iv) There is exactly one solution to the following equation

π 
ϕp t(p) = ϕp −
2

9
with respect to parameter p ∈ (A, B), which is numerically determined as

p0 = 0.010004 . . . .

For the value
π 

d0 = ϕp0 t(p0 ) = ϕp0 − = 0.0024209 . . . ,
2
the following holds
d0 = inf sup |ϕp (x)|.
p∈R+ x∈(0,π/2)

(v) For the value p0 = 0.010004 . . . , the minimax approximant of the family is deter-
mined as  3
sin x 1 4
ϕp0 (x) = − cos x + − x + 0.010004 . . . x6
x 15
and it determines the corresponding minimax approximation
 3
sin x 1
cos x − ≈ − x4 + 0.010004 . . . x6 .
x 15

Proof. This statement is based on the results of the paper [1] and the fact that

A = lim g(x) and B = lim g(x) .

x→π/2− x→0+

The function g(x) is continuous and, according to Lemma 3, strictly decreasing on

(0, π/2).
(i) If p ∈ (0, A], then

g(x) > A ⇐⇒ (A − g(x))x6 < 0 ⇐⇒ ϕA (x) < 0

and, therefore, we can conclude that

 3
1 1 sin x
x ∈ (0, π/2) =⇒ − x4 + px6 ≤ − x4 + Ax6 < cos x − .
15 15 x

(ii) If p ∈ (A, B), based on Lemma 3, the equation

g(x) = p
(p)
has a unique solution x0 and it holds
(p)
x ∈ (0, x0 ) =⇒ g(x) > p ⇐⇒ ϕp (x) < 0
 3
1 sin x
⇐⇒ − x4 + p x6 < cos x −
15 x

and
(p)
x ∈ (x0 , π/2) =⇒ g(x) < p ⇐⇒ ϕp (x) > 0
 3
sin x 1 4
⇐⇒ cos x − <− x + p x6 .
x 15

10
(iii) If p ∈ [B, ∞), then
g(x) < B ⇐⇒ (B − g(x))x6 > 0 ⇐⇒ ϕB (x) > 0
and, therefore, we can conclude that
 3
sin x 1 1
x ∈ (0, π/2) =⇒ cos x − < − x4 + Bx6 ≤ − x4 + px6 .
x 15 15
(iv), (v) Let p ∈ (A, B). For the family ϕp (x), the Taylor’s expansions are:
 
23 41 8
(2) ϕp (x) = − + p x6 + x + o(x8 )
1890 37800
and
 
8 π4 π6
ϕp (x) = − + p +
π3 240 64
 
48 π3 3 π5 π
+ − 4 +1− + p x− +
π 30 16 2
 
192 12 π2 15 π 4 π 2
+ 5
− 3
− + p x− +
π π 10 16 2
 
640 72 1 2π 5 π3 π 3
+ − 6 + 4 − − + p x− +
π π 6 15 2 2
 
1920 288 7 1 15 π 2  π 4
(3) + − 5 + 3 − + p x− +
π 7 π π 15 4 2
 
5376 960 42 1 π 5
+ − 8 + 6 − 4 + + 3πp x− +
π π π 120 2
 
14336 2880 168 61 π 6
+ 9
− 7 + 5 − 3
+p x− +
π π π 30 π 2
 
36864 8064 560 61 1 π 7
+ − 10 + 8 − 6 + 4
− x− +
π π π 5π 5040 2
 
π 7
+o x− .
2
For p ∈ (A, B), functions ϕp (x) don’t satisfy all of the conditions of the Nike theorem.
In consequence, we use the Second Nike theorem. Now we check the fulfillment of the
Second Nike theorem:
(a) Let us observe the seventh derivative
(7) d7 ϕp
ϕp (x) =
d x7
1 

= 2187 x7 − 61236 x5 + 340200 x3 − 423360 x cos3 x +
x10


+ −15309 x6 + 170100 x4 − 476280 x2 + 181440 sin x cos2 x +


+ −1641 x7 + 46116 x5 − 264600 x3 + 423360 x cos x +


+ −x10 + 3843 x6 − 44100 x4 + 158760 x2 − 181440 sin x ,

11
 (7)
for x ∈ (0, π/2). Now we prove that function ϕp (x) has exactly one zero c on (0, π/2)
(7) (7) (7)
such that ϕp (x) > 0 for x ∈ (0, c), and ϕp (x) < 0 for x ∈ (c, π/2). Function ϕp (x)
is positive for x ∈ [0, 1] because, according to [3], there exists proof that the numerator
(7)
of ϕp (x) is positive on [0, 1]. The proof is given in Appendix A2. Furthermore, let
us observe the eighth derivative

(8) d8 ϕp
ϕp (x) =
d x8
1 

= −52488 x7 + 816480 x5 − 3810240 x3 + 4354560 x cos3 x +
x11


+ −6561x8+244944x6−2041200x4+5080320x2−1814400 sin x cos2 x +


+ −x11 + 39384 x7 − 614880 x5 + 2963520 x3 − 4354560 x cos x +


+ 1641 x8 − 61488 x6 + 529200 x4 − 1693440 x2 + 1814400 sin x ,

(8)
for x ∈ (0, π/2). According to [3], there exists proof that the numerator of ϕp (x) is
(8)
negative on [1, π/2]. It is enough to prove that ϕp (x) is negative on [1, π/2] using
MacLaurin polynomials. The proof is given in Appendix A3. Finally,

−π 10 +61488 π 6 −2822400 π 4 +40642560 π 2 −185794560

ϕ(7)
p (π/2) = = −6.14789 . . . < 0 .
π 10
(7)
Therefore, there exists exactly one zero c ∈ (0, π/2) of function ϕp (x) such that
(7) (7)
ϕp (x) > 0 for x ∈ (0, c) and ϕp (x) < 0 for x ∈ (c, π/2), where c is numerically
determined as c = 1.40749 . . .. It is hereby shown that for m = 7, the first condition
of the Second Nike theorem is satisfied.

(b) According to (2), there is a right neighbourhood U0 of zero in which

ϕp (x) < 0 , ϕ′p (x) < 0 , ... , ϕ(6)
p (x) < 0 x ∈ U0

hold.

(c) According to (3), there is a left neighbourhood U π2 of π/2 in which

ϕp (x) > 0 , ϕ′p (x) > 0 , ... , ϕ(6)
p (x) > 0 x ∈ Uπ/2

hold. Then, for every function ϕp (x), on the interval (0, π/2), there exists exactly one
extremum t, which is minimum, and there exists exactly one zero x0 . Hence, for the
family of functions ϕp (x), conditions of the Second Nike theorem are satisfied, as well
as conditions of Theorem 7, which implies the existence of a minimax approximant.
Minimax approximant and error can be numerically determined via Maple software in
the following way: let f (x, p) := ϕp (x) and F (x, p) := ϕ′p (x), then using the command



fsolve {F (x, p) = 0, abs f (x, p) = f (π/2, p)}, {x = 0..π/2, p = A..B} ;

we get
{p = 0.01000418287, x = 1.299862713}

12
and, for p0 = 0.010004 . . . , we get
 3
sin x 1 4
f (x, p0 ) = − cos x + − x + 0.010004 . . . x6
x 15

and
d0 = f (π/2, p0 ) = 0.0024209 . . . .

Based on the previous considerations, enhancement of Theorem 1 has been obtained
in the following form:

Proposition 1 For every 0 < x < π/2, the following inequalities hold
 3
1 4 4π 7 − 7680 6 sin x 1 23 6
− x + x < cos x − < − x4 + x ,
15 15π 9 x 15 1890

4π 7 − 7680 23
and the constants A = = 0.0098430 . . . and B = = 0.012169 . . . are
15π 9 1890
the best possible.

Improvement of Theorem 2

Lemma 4 The family of functions

sin x cos x + 2 1 4  
ϕp (x) = − + − x + p x6 for x ∈ (0, π/2)
x 3 180
is increasingly stratified with respect to parameter p ∈ R+ .

Let us introduce the function g(x) so that the equivalence

ϕp (x) = 0 ⇐⇒ p = g(x), x ∈ (0, π/2), p ∈ R+

holds. Then
x5 − 60 x cos x + 180 sin x − 120 x
g(x) = , x ∈ (0, π/2) .
180 x7
Note that

ϕp (x) = p − g(x) x6 .

Lemma 5 The function g(x) is strictly decreasing for x ∈ (0, π/2).

Proof. Let us notice that the derivative g ′ is

270 x cos x + 30 x2 sin x − 630 sin x − x5 + 360 x

g ′ (x) = .
90 x8
It holds
g ′ (x) < 0, x ∈ (0, π/2) ⇐⇒ f (x) > 0, x ∈ (0, π/2) ,
where
f (x) = x5 − 360 x − 270 x cos x − 30 x2 sin x + 630 sin x.
According to [3], there exists proof that the mixed trigonometric polynomial function
f is positive for x ∈ (0, π/2). The proof is given in Appendix A4. 

13
Statement 2 Let

π 5 −1920π+5760 1
A= = 0.00025135 . . . and B= = 0.00026455 . . . .
45 π 7 3780
Then, it holds:
(i) If p ∈ (0, A], then

1 4 1 4 sin x cos x + 2
x ∈ (0, π/2) =⇒ − x + p x6 ≤ − x + A x6 < − .
180 180 x 3

(p)
(ii) If p ∈ (A, B), then ϕp (x) has exactly one zero x0 on (0, π/2). Also,

(p) 1 4 sin x cos x + 2

x ∈ (0, x0 ) =⇒ − x + p x6 < −
180 x 3

and
(p) sin x cos x + 2 1 4
x ∈ (x0 , π/2) =⇒ − <− x + p x6
x 3 180
hold.
(iii) If p ∈ [B, ∞), then

sin x cos x + 2 1 4 1 4
x ∈ (0, π/2) =⇒ − <− x + B x6 ≤ − x + p x6 .
x 3 180 180

(iv) There is exactly one solution to the following equation

π 
ϕp t(p) = ϕp −
2

with respect to parameter p ∈ (A, B), which is numerically determined as

p0 = 0.00025234 . . . .

For the value

π 
d0 = ϕp0 t(p0 ) = ϕp0 − = 0.000014887 . . . ,
2

the following holds

d0 = inf sup |ϕp (x)|.
p∈R+ x∈(0,π/2)

(v) For the value p0 = 0.00025234 . . . , the minimax approximant of the family is
determined as
sin x cos x + 2 1 4
ϕp0 (x) = − + − x + 0.00025234 . . . x6
x 3 180

and it determines the corresponding minimax approximation

sin x cos x + 2 1 4
− ≈− x + 0.00025234 . . . x6 .
x 3 180

14
Proof. This statement is based on the results of the paper [1] and the fact that

A = lim g(x) and B = lim g(x) .

x→π/2− x→0+

The function g(x) is continuous and, according to Lemma 5, strictly decreasing on

(0, π/2).
(i) If p ∈ (0, A], then

g(x) > A ⇐⇒ (A − g(x))x6 < 0 ⇐⇒ ϕA (x) < 0

and, therefore, we can conclude that

1 4 1 4 sin x cos x + 2
x ∈ (0, π/2) =⇒ − x + p x6 ≤ − x + A x6 < − .
180 180 x 3

(ii) If p ∈ (A, B), based on Lemma 5, the equation

g(x) = p

(p)
has a unique solution x0 and it holds

(p)
x ∈ (0, x0 ) =⇒ g(x) > p ⇐⇒ ϕp (x) < 0

1 4 sin x cos x + 2
⇐⇒ − x + p x6 < −
180 x 3

and

(p)
x ∈ (x0 , π/2) =⇒ g(x) < p ⇐⇒ ϕp (x) > 0

sin x cos x + 2 1 4
⇐⇒ − <− x + p x6 .
x 3 180

(iii) If p ∈ [B, ∞), then

g(x) < B ⇐⇒ (B − g(x))x6 > 0 ⇐⇒ ϕB (x) > 0

and, therefore, we can conclude that

sin x cos x + 2 1 4 1 4
x ∈ (0, π/2) =⇒ − <− x + B x6 ≤ − x + p x6 .
x 3 180 180

(iv), (v) Let p ∈ (A, B). For the family ϕp (x), the Taylor’s expansions are:
 
1 1
(4) ϕp (x) = − + p x6 + x8 + o(x8 )
3780 181440

15
and
 
2 2 π4 π6
ϕp (x) = − + − + p +
π 3 2880 64
 3

4 1 π 3 π5 π
+ − − + p x − +
π2 3 360 16 2
 
8 1 π2 15 π 4 π 2
+ − 3 + − + p x− +
π π 120 16 2
 
16 2 1 π 5 π3 π 3
+ 4
− 2
+ − + p x− +
π π 18 90 2 2
 
32 4 1 1 15 π 2 π 4
(5) + − 5 + 3 − − + p x− +
π π 12 π 180 4 2
 
64 8 1 1 π 5
+ 6
− 4
+ 2
− + 3 π p x− +
π π 6π 360 2
 
128 16 1 1 π 6
+ − 7 + 5 − 3
+ + p x− +
π π 3π 360 π 2
 
256 32 2 1 1 π 7
+ 8
− 6
+ 4
− 2
+ x− +
π π 3π 180 π 15120 2
 
π 7
+o x− .
2

For p ∈ (A, B), functions ϕp (x) satisfy all of the conditions of the Nike theorem. Now
we check the fulfillment of the Nike theorem:
(a) Let us observe the seventh derivative

(7) 1  8

ϕp (x) = x − 21 x6 + 630 x4 − 7560 x2 + 15120 sin x +
3x8


+ 3 x x6 − 42 x4 + 840 x2 − 5040 cos x

(7)
for x ∈ (0, π/2). According to [3], there exists proof that the numerator of ϕp (x) is
positive on [0, π/2]. The proof is given in Appendix A5. Thus, for m = 7, the first
condition of the Nike theorem is satisfied.
(b) According to (4), there is a right neighbourhood U0 of zero in which


ϕp (x) < 0 , ϕ′p (x) < 0 , ... , ϕ(6)
p (x) < 0 x ∈ U0

hold.
(c) According to (5), there is a left neighbourhood U π2 of π/2 in which


ϕp (x) > 0 , ϕ′p (x) > 0 , ... , ϕ(6)
p (x) > 0 x ∈ Uπ/2

hold. Then, for every function ϕp (x), on the interval (0, π/2), there exists exactly
one extremum t, which is minimum, and there exists exactly one zero x0 . Hence, for
the family of functions ϕp (x), conditions of the Nike theorem are satisfied, as well
as conditions of Theorem 7, which implies the existence of a minimax approximant.

16
Minimax approximant and error can be numerically determined via Maple software in
the following way: let f (x, p) := ϕp (x) and F (x, p) := ϕ′p (x), then using the command



fsolve {F (x, p) = 0, abs f (x, p) = f (π/2, p)}, {x = 0..π/2, p = A..B} ;

we get
{p = 0.000252341144, x = 1.305655179}
and, for p0 = 0.00025234 . . . , we get

sin x cos x + 2 1 4
f (x, p0 ) = − + − x + 0.00025234 . . . x6
x 3 180

and
d0 = f (π/2, p0 ) = 0.000014887 . . . .


Based on the previous considerations, enhancement of Theorem 2 has been obtained

in the following form:

Proposition 2 For every 0 < x < π/2, the following inequalities hold

1 4 π 5 − 1920π + 5760 6 sin x cos x + 2 1 4 1

− x + x < − <− x + x6 ,
180 45 π 7 x 3 180 3780

π 5 − 1920π + 5760 1
and the constants A = = 0.00025135 . . . and B = = 0.00026455 . . .
45 π 7 3780
are the best possible.

Improvement of Theorem 3

Lemma 6 The family of functions

sin x tan x 3 4 1 1  
ϕp (x) = 2 + −3− x + p x6 for x ∈ (0, π/2)
x x 20 cos x cos x

is increasingly stratified with respect to parameter p ∈ R+ .

Let us introduce the function g(x) so that the equivalence

ϕp (x) = 0 ⇐⇒ p = g(x), x ∈ (0, π/2), p ∈ R+

holds. Then

3x5 − 40 sin x cos x + 60 x cos x − 20 sin x

g(x) = , x ∈ (0, π/2) .
20 x7

Note that

x6
ϕp (x) = p − g(x) .
cos x

Lemma 7 The function g(x) is strictly decreasing for x ∈ (0, π/2).

17
Proof. Let us notice that the derivative g ′ is


′ −40 x cos2 x − 190 x cos x + 140 sin x cos x − 30 x2 − 70 sin x − 3x5 + 20 x
g (x) = .
10 x8
It holds
g ′ (x) < 0, x ∈ (0, π/2) ⇐⇒ f (x) > 0, x ∈ (0, π/2) ,
where


f (x) = 40 x cos2 x + 190 x cos x − 140 sin x cos x + 30 x2 − 70 sin x + 3x5 − 20 x.

According to [3], there exists proof that the mixed trigonometric polynomial function
f is positive for x ∈ (0, π/2). The proof is given in Appendix A6. 

Statement 3 Let
3π 5 − 640 3
A= = 0.018412 . . . and B= = 0.021428 . . . .
5π 7 140
Then, it holds:
(i) If p ∈ (0, A], then

sin x tan x 3 4 1 1
x ∈ (0, π/2) =⇒ 2 + <3+ x − A x6 ≤
x x 20 cos x cos x
3 4 1 1
≤3+ x − p x6 .
20 cos x cos x
(p)
(ii) If p ∈ (A, B), then ϕp (x) has exactly one zero x0 on (0, π/2). Also,

(p) sin x tan x 3 4 1 1

x ∈ (0, x0 ) =⇒ 2 + <3+ x − p x6
x x 20 cos x cos x
and
(p) 3 4 1 1 sin x tan x
x ∈ (x0 , π/2) =⇒ 3 + x − p x6 <2 +
20 cos x cos x x x
hold.
(iii) If p ∈ [B, ∞), then
3 4 1 1 3 4 1 1
x ∈ (0, π/2) =⇒ 3+ x − p x6 ≤3+ x − B x6
20 cos x cos x 20 cos x cos x
sin x tan x
<2 + .
x x

Proof. This statement is based on the results of the paper [1] and the fact that

A = lim g(x) and B = lim g(x) .

x→π/2− x→0+

The function g(x) is continuous and, according to Lemma 7, strictly decreasing on

(0, π/2).
(i) If p ∈ (0, A], then

x6
g(x) > A ⇐⇒ (A − g(x)) < 0 ⇐⇒ ϕA (x) < 0
cos x

18
and, therefore, we can conclude that

sin x tan x 3 4 1 1
x ∈ (0, π/2) =⇒ 2 + <3+ x − A x6 ≤
x x 20 cos x cos x
3 4 1 1
≤3+ x − p x6 .
20 cos x cos x

(ii) If p ∈ (A, B), based on Lemma 7, the equation

g(x) = p
(p)
has a unique solution x0 and it holds
(p)
x ∈ (0, x0 ) =⇒ g(x) > p ⇐⇒ ϕp (x) < 0

sin x tan x 3 4 1 1
⇐⇒ 2 + <3+ x − p x6
x x 20 cos x cos x
and
(p)
x ∈ (x0 , π/2) =⇒ g(x) < p ⇐⇒ ϕp (x) > 0

3 4 1 1 sin x tan x
⇐⇒ 3 + x − p x6 <2 + .
20 cos x cos x x x

(iii) If p ∈ [B, ∞), then

x6
g(x) < B ⇐⇒ (B − g(x)) > 0 ⇐⇒ ϕB (x) > 0
cos x
and, therefore, we can conclude that
3 4 1 1 3 4 1 1
x ∈ (0, π/2) =⇒ 3+ x − p x6 ≤3+ x − B x6
20 cos x cos x 20 cos x cos x
sin x tan x
<2 + .
x x

Let us notice that ϕB (π/2−) = +∞. Hence, one of the conditions of Theorem 7 is
not satisfied, thus, the minimax approximant is not considered.
Based on the previous considerations, enhancement of Theorem 3 has been obtained
in the following form:

Proposition 3 For every 0 < x < π/2, the following inequalities hold
   
3x4 3x6 1 sin x tan x 3x4 (3π 5 − 640)x6 1
3+ − < 2 + < 3+ − ,
20 140 cos x x x 20 5π 7 cos x

3π 5 − 640 3
and the constants A = = 0.018412 . . . and B = = 0.021428 . . . are the
5π 7 140
best possible.

19
Improvement of Theorem 4

Lemma 8 The family of functions

 2  
sin x tan x 8 4 1 1
ϕp (x) = + −2− x + p x6 for x ∈ (0, π/2)
x x 45 cos x cos x

is increasingly stratified with respect to parameter p ∈ R+ .

Let us introduce the function g(x) so that the equivalence

ϕp (x) = 0 ⇐⇒ p = g(x), x ∈ (0, π/2), p ∈ R+

holds. Then
8x6 + 90 x2 cos x − 45 sin2 x cos x − 45 x sin x
g(x) = , x ∈ (0, π/2) .
45 x8
Note that

x6
ϕp (x) = p − g(x) .
cos x
Lemma 9 The function g(x) is strictly decreasing for x ∈ (0, π/2).

Proof. Let us notice that the derivative g ′ is

′ −360 cos3 x−135 x sin x cos2 x− 585 x2 −360 cos x− 90 x3 −360x sin x−16 x6
g (x) = .
45 x9
It holds
g ′ (x) < 0, x ∈ (0, π/2) ⇐⇒ f (x) > 0, x ∈ (0, π/2) ,
where



f (x) = 360 cos3 x + 135x sin x cos2 x + 585 x2 − 360 cos x + 90x3 − 360x sin x + 16x6 .

According to [3], there exists proof that the mixed trigonometric polynomial function
f is positive for x ∈ (0, π/2). The proof is given in Appendix A7. 

Statement 4 Let
32π 5 − 5760 4
A= = 0.029670 . . . and B= = 0.038095 . . . .
45π 7 105
Then, it holds:
(i) If p ∈ (0, A], then
 2
sin x
tan x 8 4 1 1
x ∈ (0, π/2) =⇒ <2++ x − A x6 ≤
x x 45 cos x cos x
8 4 1 1
≤2+ x − p x6 .
45 cos x cos x
(p)
(ii) If p ∈ (A, B), then ϕp (x) has exactly one zero x0 on (0, π/2). Also,
 sin x 2 tan x 8 4 1 1
(p)
x ∈ (0, x0 ) =⇒ + <2+ x − p x6
x x 45 cos x cos x

20
and  sin x 2 tan x
(p) 8 4 1 1
x ∈ (x0 , π/2) =⇒ 2 + x − p x6 < +
45 cos x cos x x x
hold.
(iii) If p ∈ [B, ∞), then

8 4 1 1 8 4 1 1
x ∈ (0, π/2) =⇒ 2+ x − p x6 ≤2+ x − B x6
45 cos x cos x 45 cos x cos x
 sin x 2 tan x
< + .
x x

Proof. This statement is based on the results of the paper [1] and the fact that

A = lim g(x) and B = lim g(x) .

x→π/2− x→0+

The function g(x) is continuous and, according to Lemma 9, strictly decreasing on

(0, π/2).
(i) If p ∈ (0, A], then

x6
g(x) > A ⇐⇒ (A − g(x)) < 0 ⇐⇒ ϕA (x) < 0
cos x
and, therefore, we can conclude that
 2
tan x
sin x 8 4 1 1
x ∈ (0, π/2) =⇒ <2++ x − A x6 ≤
x x 45 cos x cos x
8 4 1 1
≤2+ x − p x6 .
45 cos x cos x

(ii) If p ∈ (A, B), based on Lemma 9, the equation

g(x) = p
(p)
has a unique solution x0 and it holds
(p)
x ∈ (0, x0 ) =⇒ g(x) > p ⇐⇒ ϕp (x) < 0
 sin x 2 tan x 8 4 1 1
⇐⇒ + <2+ x − p x6
x x 45 cos x cos x
and
(p)
x ∈ (x0 , π/2) =⇒ g(x) < p ⇐⇒ ϕp (x) > 0

8 4 1 1  sin x 2 tan x
⇐⇒ 2 + x − p x6 < + .
45 cos x cos x x x

(iii) If p ∈ [B, ∞), then

x6
g(x) < B ⇐⇒ (B − g(x)) = > 0 ⇐⇒ ϕB (x) > 0
cos x

21
and, therefore, we can conclude that
8 4 1 1 8 4 1 1
x ∈ (0, π/2) =⇒ 2+ x − p x6 ≤2+ x − B x6
45 cos x cos x 45 cos x cos x
 sin x 2 tan x
< + .
x x

Let us notice that ϕB (π/2−) = +∞. Hence, one of the conditions of Theorem 7 is
not satisfied, thus, the minimax approximant is not considered.
Based on the previous considerations, enhancement of Theorem 4 has been obtained
in the following form:

Proposition 4 For every 0 < x < π/2, the following inequalities hold
   2  
8x4 4x6 1 sin x tan x 8x4 (32π 5 − 5760)x6 1
2+ − < + < 2+ − ,
45 105 cos x x x 45 45π 7 cos x

32π 5 − 5760 4
and the constants A = = 0.029670 . . . and B = = 0.038095 . . . are
45π 7 105
the best possible.

Analogously to Statement 1 and Statement 2, the following statements can be proved:

Improvement of Theorem 5

Lemma 10 The family of functions

 x 2 x  
ϕp (x) = + − 2 − p x4 for x ∈ (0, π/2)
sin x tan x
is decreasingly stratified with respect to parameter p ∈ R+ .

Statement 5 Let
2 4π 2 − 32
A= = 0.044444 . . . and B= = 0.076773 . . . .
45 π4
Then, it holds:
(i) If p ∈ (0, A], then
 x 2 x
x ∈ (0, π/2) =⇒ 2 + px4 ≤ 2 + Ax4 < + .
sin x tan x
(p)
(ii) If p ∈ (A, B), then ϕp (x) has exactly one zero x0 on (0, π/2). Also,
 x 2 x
(p)
x ∈ (0, x0 ) =⇒ + < 2 + p x4
sin x tan x
and  x 2
(p) x
x ∈ (x0 , π/2) =⇒ 2 + p x4 < +
sin x tan x
hold.

22
(iii) If p ∈ [B, ∞), then
 x 2 x
x ∈ (0, π/2) =⇒ + < 2 + Bx4 ≤ 2 + px4 .
sin x tan x

(iv) There is exactly one solution to the following equation

π 
ϕp t(p) = ϕp −
2
with respect to parameter p ∈ (A, B), which is numerically determined as

p0 = 0.072425 . . . .

For the value
π 

d0 = ϕp0 t(p0 ) = ϕp0 − = 0.026471 . . . ,
2
the following holds
d0 = inf sup |ϕp (x)|.
p∈R+ x∈(0,π/2)

(v) For the value p0 = 0.072425 . . . , the minimax approximant of the family is deter-
mined as  x 2 x
ϕp0 (x) = + − 2 − 0.072425 . . . x4
sin x tan x
and it determines the corresponding minimax approximation
 x 2 x
+ ≈ 2 + 0.072425 . . . x4 .
sin x tan x
Based on the previous considerations, enhancement of Theorem 5 has been obtained
in the following form:

Proposition 5 For every 0 < x < π/2, the following inequalities hold

2 4  x 2 x 4π 2 − 32 4
2+ x < + < 2+ x ,
45 sin x tan x π4
2 4π 2 − 32
and the constants A = = 0.044444 . . . and B = = 0.076773 . . . are the
45 π4
best possible.

Improvement of Theorem 6

Lemma 11 The family of functions

x 1 4  
ϕp (x) = 3 + cos x − 4 − x − p x6 for x ∈ (0, π/2)
sin x 10
is decreasingly stratified with respect to parameter p ∈ R+ .

Statement 6 Let
1 480π−2π 4 −1280
A= = 0.0047619 . . . and B= = 0.0068954 . . . .
210 5π 6

23
Then, it holds:
(i) If p ∈ (0, A], then
1 4 1 4 x
x ∈ (0, π/2) =⇒ 4 + x + px6 ≤ 4 + x + Ax6 < 3 + cos x.
10 10 sin x
(p)
(ii) If p ∈ (A, B), then ϕp (x) has exactly one zero x0 on (0, π/2). Also,
(p) x 1 4
x ∈ (0, x0 ) =⇒ 3 + cos x < 4 + x + px6
sin x 10
and
(p) 1 4 x
x ∈ (x0 , π/2) =⇒ 4 + x + px6 < 3 + cos x
10 sin x
hold.
(iii) If p ∈ [B, ∞), then
x 1 4 1 4
x ∈ (0, π/2) =⇒ 3 + cos x < 4 + x + Bx6 ≤ 4 + x + px6 .
sin x 10 10
(iv) There is exactly one solution to the following equation

π 
ϕp t(p) = ϕp −
2
with respect to parameter p ∈ (A, B), which is numerically determined as
p0 = 0.0066982 . . . .
For the value
π 
d0 = ϕp0 t(p0 ) = ϕp0 − = 0.0029637 . . . ,
2
the following holds
d0 = inf sup |ϕp (x)|.
p∈R+ x∈(0,π/2)

(v) For the value p0 = 0.0066982 . . . , the minimax approximant of the family is
determined as
x 1 4
ϕp0 (x) = 3 + cos x − 4 − x − 0.0066982 . . . x6
sin x 10
and it determines the corresponding minimax approximation
x 1 4
3 + cos x ≈ 4 + x + 0.0066982 . . . x6 .
sin x 10
Based on the previous considerations, enhancement of Theorem 6 has been obtained
in the following form:

Proposition 6 For every 0 < x < π/2, the following inequalities hold
1 4 1 6 x 1 4 480π − 2π 4 − 1280 6
4+ x + x < 3 + cos x < 4 + x + x ,
10 210 sin x 10 5π 6
1 480π − 2π 4 − 1280
and the constants A = = 0.0047619 . . . and B = = 0.0068954 . . .
210 5π 6
are the best possible.

The existence of minimax approximant in Statement 5 and Statement 6 is a conse-

quence of Theorem 7’ and Theorem 8.

24
4 Conclusion
This paper specifies the results of C. Mortici [1] using the method described in [2].
The mentioned method could be applied for possible improvements of existing results
from the Theory of analytic inequalities [12], [13], [28] in terms of determining the
corresponding minimax approximants for various inequalities. In the previous section,
examples of minimax approximations are presented where they exist. It is important
to note that through minimax approximants, the error in approximations is minimized
in the considered sense.
The main aim of this paper is to promote SimTheP, an automated theorem prover for
MTP inequalities, developed through the doctoral dissertation [15]. All the essential
proofs of MTP inequalities in this paper are given in the Appendix and derived using
the prover SimTheP. For any given MTP inequality f (x) > 0, for x ∈ S ⊆ [0, π/2],
SimTheP provides a structured proof divided into parts I-IV. Each part is designed to
allow manual step-by-step verification, demonstrating SimTheP’s capability to repli-
cate the human way of proving inequalities.
Many results within the Theory of analytic inequalities, for example, results from
the papers [7], [8], [29]-[68], could be proved using the prover SimTheP, see the link
https://simthep.etf.bg.ac.rs/.
It is crucial to highlight that through Statements 1–6, all Theorems 1–6 have been
improved and minimax approximations have been determined wherever feasible. As a
result, Propositions 1–6 were obtained, where for the inequalities considered in Theo-
rems 1–6, the best possible constants were identified. Such an approach to Theorems
1–6 was enabled by the utilization of the concept of stratification. Moreover, this
paper presents the first paper in which the automated theorem prover SimTheP was
utilized to deliver proofs for the MTP inequalities, marking a significant advancement
in the field.

Acknowledgments
The authors are greatly indebted to Dr. Ivana Jovović for numerous stimulating
conversations about the concept of stratification. Her assistance in applying MTP
inequalities to various problems, coupled with her ongoing and steadfast commitment
to supporting our endeavors, has been instrumental and is deeply appreciated. Special
appreciation is also extended to Dr. Marija Nenezić Jović for her invaluable comments
that greatly enhanced certain proofs. The authors also express their gratitude to the
referees for their thorough reading of the paper and their valuable suggestions and
comments.
The work of the first author was financially supported by the Ministry of Science,
Technological Development and Innovation of the Republic of Serbia under contract
number: 451-03-65/2024-03/200156.
The work of the second, fourth and fifth authors was financially supported by the
Ministry of Science, Technological Development and Innovation of the Republic of
Serbia under contract number: 451-03-65/2024-03/200103.
The work of the third author was financially supported by the Ministry of Science,
Technological Development and Innovation of the Republic of Serbia under contract
number: 451-03-66/2024-03/200103.

25
References
[1] C. Mortici: The natural approach of Wilker-Cusa-Huygens inequalities, Mathe-
matical Inequalities & Applications 14:3 (2011), 535–541.
[2] B. Malešević, B. Mihailović: A Minimax Approximant in the Theory of An-
alytical Inequalities, Applicable Analysis and Discrete Mathematics 15:2 (2021),
486–509.
[3] B. Malešević, M. Makragić: A Method for Proving Some Inequali-
ties on Mixed Trigonometric Polynomial Functions, Journal of Mathemat-
ical Inequalities 10:3 (2016), 849–876 (see also and preprint versions at
https://arxiv.org/abs/1504.08345)
[4] B. Malešević, T. Lutovac, B. Banjac: One Method for Proving Some Classes
of Exponential Analytical Inequalities, Filomat 32:20 (2018), 6921–6925.
[5] B. Malešević, B. Mihailović, M. Nenezić Jović, L. Milinković: Some mini-
max approximants of D’Aurizio trigonometric inequalities, HAL (Preprint) (2022),
1–9.
[6] B. Malešević, M. Mićović: Exponential Polynomials and Stratification in the
Theory of Analytic Inequalities, Journal of Science and Arts 23:3 (2023), 659–670.
[7] M. Mićović, B. Malešević: Jordan-Type Inequalities and Stratification, Re-
searchGate (Preprint) (2024), 1–30.
[8] B. Malešević, D. Jovanović: Frame’s Types of Inequalities and Stratification,
CUBO, A Mathematical Journal 26:1 (2024), 1–15.
[9] B. Yu, B. Dong: A Hybrid Polynomial System Solving Method for Mixed Trigono-
metric Polynomial Systems, SIAM Journal on Numerical Analysis, 46:3 (2008)
1503–1518.
[10] S. Chen, Z. Liu: Automated proving of trigonometric function inequalities using
Taylor expansion, Journal of Systems Science and Mathematical Sciences, 36:8
(2016), 1339-1348. (in Chinese)
[11] S. Chen, Z. Liu: Automated proof of mixed trigonometric-polynomial inequalities,
Journal of Symbolic Computation, 101 (2020), 318-329.
[12] D. S. Mitrinović: Analytic inequalities, Springer-Verlag, 1970.
[13] G. Milovanović, M. Rassias (editors): Analytic Number Theory, Approxi-
mation Theory and Special functions, Springer 2014, Chapter: G.D. Anderson, M.
Vuorinen, X. Zhang: Topics in Special Functions III, 297–345.
[14] M. Makragić: On trigonometric polynomial ring with applications in
the theory of analytic inequalities, Faculty of Mathematics, Belgrade
2018, Ph.D. thesis in Serbian, see link of National Repository of Dis-
sertations in Serbia https://nardus.mpn.gov.rs/ and mathgenealogy link
https://www.mathgenealogy.org/id.php?id=239436
[15] B. Banjac: System for automatic proving of some classes of analytic inequalities,
School of Electrical Engineering, Belgrade 2019, Ph.D. thesis in Serbian, see link
of National Repository of Dissertations in Serbia https://nardus.mpn.gov.rs/ and
mathgenealogy link https://www.mathgenealogy.org/id.php?id=248798

26
[16] M. Nenezić Jović: Stratified Families of Functions in the Theory of An-
alytical Inequalities With Applications, School of Electrical Engineering, Bel-
grade 2023, Ph.D. thesis in Serbian, see link of National Repository of
Dissertations in Serbia https://nardus.mpn.gov.rs/ and mathgenealogy link
https://www.mathgenealogy.org/id.php?id=307785
[17] B. Malešević: One Method for Proving Inequalities by Computer, Journal of
Inequalities and Applications (2007), 1–8.
[18] B. Banjac, M. Makragić, B. Malešević: Some Notes on a Method for Proving
Inequalities by Computer, Results in Mathematics 69 (2016), 161–176.
[19] B. Malešević, B. Banjac: Automated Proving Mixed Trigonometric Polynomial
Inequalities, Proceedings of 27th TELFOR conference, Serbia, Belgrade, November
26-27, 2019.
[20] B. Malešević, B. Banjac: One method for proving polynomial inequalities
with real coefficients, Proceedings of 28th TELFOR conference, Serbia, Belgrade,
November 24-25, 2020.
[21] N. Cutland: Computalibity - an introduction to recursive funtion theory, Cam-
bridge University Press 1980.
[22] J.C.F. Sturm: Mémoire sur la résolution des équations numériques, Bulletin des
Sciences de Ferussac 11 (1829), 419–425.
[23] B. Malešević, B. Banjac, V. Šešum Čavić, N. Korolija: One algorithm for
testing annulling of mixed trigonometric polynomial functions on boundary points,
Proceedings of 30th TELFOR conference, Serbia, Belgrade, November 15-16, 2022.
[24] H. Alzer, M. K. Kwong: A refinement of Vietoris’ inequality for cosine poly-
nomials, Analysis and Applications 14:5 (2016), 615–629.
[25] H. Alzer, M. K. Kwong: On two trigonometric inequalities of Askey and
Steinig, New York Journal of Mathematics 26 (2020), 28–36.
[26] H. Alzer, M. K. Kwong: On Fejér’s inequalities for the Legendre polynomials,
Mathematische Nachrichten 290:2 (2017), 1–15.
[27] H. Alzer, M. K. Kwong: Inequalities for trigonometric sums, The Journal of
Analysis (2024)
[28] M. J. Cloud, B. C. Drachman, L. P. Lebedev: Inequalities with Applications
to Engineering, Springer 2014.
[29] S. Chen, X. Ge: A solution to an open problem for Wilker-type inequalities,
Journal of Mathematical Inequalities 15:1 (2021), 59–65.
[30] F. Qi, D.-W. Niu, B.-N. Guo: Refinements, Generalizations, and Applications of
Jordan’s Inequality and Related Problems, Journal of Inequalities and Applications
2009 (2009), 1–52.
[31] B. A. Bhayo, J. Sándor: On classical inequalities of trigonometric and hyper-
bolic functions, arXiv (Preprint) (2014), 1–59.
[32] M. Nenezić, B. Malešević, C. Mortici: New approximations of some expres-
sions involving trigonometric functions, Applied Mathematics and Computation
283 (2016), 299–315.
[33] B. Malešević, M. Nenezić, L. Zhu, B. Banjac, M. Petrović: Some new
estimates of precision of Cusa-Huygens and Huygens approximations, Applicable
Analysis and Discrete Mathematics 15:1 (2021), 243–259.

27
[34] L. Zhu, M. Nenezić: New approximation inequalities for circular functions, Jour-
nal of Inequalities and Applications 2018:313 (2018), 1–12.
[35] B. Malešević, B. Banjac, I. Jovović: A proof of two conjectures of Chao-Ping
Chen for inverse trigonometric functions, Journal of Mathematical Inequalities
11:1 (2017), 151–162.
[36] B. Malešević, T. Lutovac, B. Banjac: A proof of an open problem of Yusuke
Nishizawa for a power-exponential function, Journal of Mathematical Inequalities
12:2 (2018), 473–485.
[37] T. Lutovac, B. Malešević, C. Mortici: The natural algorithmic approach of
mixed trigonometric-polynomial problems, Journal of Inequalities and Applications
2017:116 (2017), 1–16.
[38] B. Malešević, T. Lutovac, M. Rašajski, C. Mortici: Extensions of the natu-
ral approach to refinements and generalizations of some trigonometric inequalities,
Advances in Difference Equations 2018:90 (2018), 1–15.
[39] C.-P. Chen, B. Malešević: Sharp inequalities related to the Adamović-
Mitrinović, Cusa, Wilker and Huygens results, Filomat 37:19 (2023), 6319–6334.
[40] Y. J. Bagul, B. Banjac, C. Chesneau, M. Kostić, B. Malešević: New
Refinements of Cusa-Huygens Inequality, Results in Mathematics 76:107 (2021),
1–16.
[41] L. Zhu, B. Malešević: New inequalities of Huygens-type involving tangent and
sine functions, Hacettepe Journal of Mathematics and Statistics 52:1 (2023), 36–
61.
[42] Y. J. Bagul, C. Chesneau, M. Kostić: On the Cusa-Huygens inequality, Re-
vista de la Real Academia de Ciencias Exactas, Fı́sicas y Naturales. Serie A.
Matemáticas 115:29 (2021), 1–12.
[43] A. R. Chouikha: New sharp inequalities related to classical trigonometric in-
equalities, Journal of Inequalities and Special Functions 11:4 (2020), 27–35.
[44] Y. Hu, C. Mortici: A Lower Bound on the Sinc Function and Its Application,
The Scientific World Journal 2014 (2014), 1–4.
[45] L. Debnath, C. Mortici, L. Zhu: Refinements of Jordan-Stečkin and Becker-
Stark Inequalities, Results in Mathematics 67 (2015), 207–215.
[46] R. Shinde, C. Chesneau, N. Darkunde, S. Ghodechor, A. Lagad: Revisit
of an Improved Wilker Type Inequality, Pan-American Journal of Mathematics 2
(2023), 1-17.
[47] Y. J. Bagul, C. Chesneau: Refined forms of Oppenheim and Cusa-Huygens type
inequalities, Acta et Commentationes Universitatis Tartuensis de Mathematica
24:2 (2020), 183–194.
[48] Y. J. Bagul, C. Chesneau, M. Kostić: The Cusa-Huygens inequality revisited,
Novi Sad Journal of Mathematics 50:2 (2020), 149–159.
[49] Y. J. Bagul, C. Chesneau: Generalized Bounds for Sine And Cosine Functions,
Asian-European Journal of Mathematics 15:1 (2022), 1–16.
[50] R. M. Dhaigude, C. Chesneau, Y. J. Bagul: About Trigonometric-polynomial
Bounds of Sinc Function, Mathematical Sciences and Applications E-Notes 8:1
(2020), 100–104.

28
[51] Y. J. Bagul, R. M. Dhaigude, M. Kostić, C. Chesneau: Polynomial-
Exponential Bounds for Some Trigonometric and Hyperbolic Functions, Axioms
10:4 (2021), 1–10.
[52] B. Zhang, C.-P. Chen: Sharp Wilker and Huygens type inequalities for trigono-
metric and inverse trigonometric functions, Journal of Mathematical Inequalities
14:3 (2020), 673–684.
[53] C.-P. Chen, R. B. Paris: On the Wilker and Huygens-type inequalities, Journal
of Mathematical Inequalities 14:3 (2020), 685–705.
[54] B. Zhang, C.-P. Chen: Sharpness and generalization of Jordan, Becker-Stark
and Papenfuss inequalities with an application, Journal of Mathematical Inequali-
ties 13:4 (2019), 1209–1234.
[55] C.-P. Chen, N. Elezović: Sharp Redheffer-type and Becker-Stark-type inequal-
ities with an application, Mathematical Inequalities & Applications 21:4 (2018),
1059–1078.
[56] Q.-X. Qiao, C.-P. Chen: Approximations to inverse tangent function, Journal
of Inequalities and Applications 2018:141 (2018), 1–14.
[57] C.-P. Chen, F. Qi: Inequalities of some trigonometric functions, Publikacije
Elektrotehničkog fakulteta. Serija Matematika 15 (2004), 72–79.
[58] B.-N. Guo, Q.-M. Luo, F. Qi: Sharpening and generalizations of Shafer-Fink’s
double inequality for the arc sine function, Filomat 27:2 (2013), 261–265.
[59] W.-D. Jiang, Q.-M. Luo, F. Qi: Refinements and Sharpening of some Huygens
and Wilker Type Inequalities, Turkish Journal of Analysis and Number Theory 2:4
(2014), 134–139.
[60] G. Bercu: The natural approach of trigonometric inequalities - Pade approxi-
mant, Journal of Mathematical Inequalities 11:1 (2017), 181–191.
[61] W.-D. Jiang: New sharp inequalities of Mitrinović-Adamović type, Applicable
Analysis and Discrete Mathematics 17:1 (2023), 76–91.
[62] C.-P. Chen, C. Mortici: The relationship between Huygens’ and Wilker’s in-
equalities and further remarks, Applicable Analysis and Discrete Mathematics 17:1
(2023), 92–100.
[63] L. Zhu, Z. Sun: Refinements of Huygens- and Wilker- type inequalities, AIMS
Mathematics 5:4 (2020), 2967–2978.
[64] L. Zhu: New inequalities of Wilker’s type for circular functions, AIMS Mathe-
matics 5:5 (2020), 4874–4888.
[65] L. Zhu: New Inequalities of Cusa-Huygens Type, Mathematics 9:17 (2021), 1–13.
[66] A. R. Chouikha, C. Chesneau, Y. J. Bagul: Some refinements of well-known
inequalities involving trigonometric functions, Journal of the Ramanujan Mathe-
matical Society 36:3 (2021), 193–202.
[67] Y. J. Bagul, S. B. Thool, C. Chesneau, R. M. Dhaigude: Refinements of
some classical inequalities involving sinc and hyperbolic sinc functions, Annales
Mathematicae Silesianae 37:1 (2023), 1–15.
[68] G. Bercu: Refinements of Wilker-Huygens-Type Inequalities via Trigonometric
Series, Symmetry 13:8 (2021), 1–13.

29
Appendix
This Appendix was created using the automated prover SimTheP, which for the MTP
function and the segment gives as output Tex/pdf files that were directly transferred
to the Appendix.

Appendix A1
The initial MTP function is


f (x) = 2 · x7 + 135 · sin x · cos2 x + 15 · x4 − 135 · sin x


−45 · x · cos3 x + 90 · x3 + 45 · x · cos x


and the initial interval is S = 0, π2 .


Automated proof that f (x) > 0 for x ∈ 0, π2 :


I (Recognition of possible case) Facts f (0) = 0 and
f π2 6= 0 are correct.

The MTP
function f is positive at boundary
point π2 f π2 = 3.51310 . . . > 0 . Therefore, it
is possible that f (x) > 0 over 0, π2 .
II (Transformation of angles) After the transformation of terms cosm x · sinn x (m, n ∈
N0 ) into the sum of sine and cosine functions of multiple angles, in the MTP function
f , we obtain


f (x) = − 45
4
· x · cos 3x + 15 · x4 − 405
4
· sin x
3

7
+ 90 · x + 4 · x · cos x + 2 · x + 135
45
4
· sin 3x .

Then, we consider the previous expression as two separate expressions, the first with
positive and the second with negative terms next to sine and cosine functions


f + (x) = 135
4
· sin 3x + 90 · x3 + 45
4
· x · cos x + 15 · x4 · sin x + 2 · x7 ,

f − (x) = − 405
4
· sin x − 45
4
· x · cos 3x .

III (Determination of downward rational polynomial approximation) After substitu-

tion of sine and cosine functions by appropriate (downward or upward) polynomial
approximations, we obtain downward polynomial approximations of f + (x) and f − (x)
respectively
sin,0


Pi+0 ,i1 ,i2 (x) = 135
4
· T 4·i0 +3
(3x) + 90 · x3 + 45
4
· x · T cos,0
4·i1 +2 (x)

+15 · x4 · T sin,0 7
4·i2 +3 (x) + 2 · x ,

sin,0 cos,0
Pi−3 ,i4 (x) = − 405
4
· T 4·i3 +1 (x) − 45
4
· x · T 4·i4 (3x) .

For concrete indices (i0 , i1 , i2 , i3 , i4 ) = (2, 2, 1, 2, 3), we obtain

sin,0

+
P2,2,1 (x) = 135 4
· T 4·2+3 (3x) + 90 · x3 + 45 4
· x · T cos,0
4·2+2 (x)

+15 · x4 · T sin,0 7
4·1+3 (x) + 2 · x ,

+
P2,2,1 1
(x) = − 40320 · x13 − 133523
887040
· x11 + 3281
1792
· x9
− 639
56
· x7 + 621
16
· x5 − 135
2
· x3 + 225
2
·x

30
and
sin,0 cos,0
−
P2,3 (x) = − 405
4
· T 4·2+1 (x) − 45
4
· x · T 4·3 (3x) ,
−
P2,3 19683
(x) = − 1576960 · x13 + 6561
35840
· x11 − 3281
1792
· x9
+ 639
56
· x7 − 621
16
· x5 + 1352
· x3 − 225
2
· x.
Finally, for the MTP function

f (x) = f + (x) + f − (x)

we obtain the concrete downward polynomial approximation

+ −
P (x) = P2,2,1 (x) + P2,3 (x)
25357
= − 2027520 · x13 + 115447
3548160
· x11
π


over 0, 2
, i.e. it holds that
f (x) > P (x)

π
over 0, . 2

IV (The final part) Based on the Sturm theorem, the following inequality

P (x) > 0


is true over 0, π2 . The stated conclusion for the polynomial function P is correct
based on the following facts:
1. We can conclude, by Sturm theorem, that the polynomial function P (x) has only

one zero over the concrete extended segment [−0.1, 1.58] of the initial interval 0, π2 .
2. Facts P (−0.1) 6= 0, P (0) = 0 and P (1.58) 6= 0 are correct.



3. The polynomial P is positive at boundary point π2 P π2 = 0.24116 . . . > 0 .
Therefore, the following inequality

f (x) > 0
π


is true over 0, 2
. 

Appendix A2
The initial MTP function is


f (x) = −15309 · x6 + 170100 · x4 − 476280 · x2 + 181440 · sin x · cos2 x


+ −x10 + 3843 · x6 − 44100 · x4 + 158760 · x2 − 181440 · sin x


+ 2187 · x7 − 61236 · x5 + 340200 · x3 − 423360 · x · cos3 x


+ −1641 · x7 + 46116 · x5 − 264600 · x3 + 423360 · x · cos x

and the initial interval is S = (0, 1].

Automated proof that f (x) > 0 for x ∈ (0, 1]:
I (Recognition of possible case) Facts f (0) = 0 and f (1) 6= 0 are correct.

The MTP
function f is positive at boundary point 1 f (1) = 14.68957 . . . > 0 . Therefore, it
is possible that f (x) > 0 over (0, 1].

31
II (Transformation of angles) After the transformation of terms cosm x · sinn x (m, n ∈
N0 ) into the sum of sine and cosine functions of multiple angles, in the MTP function
f , we obtain


f (x) = − 15309
4
· x6 + 42525 · x4 − 119070 · x2 + 45360 · sin 3x


+ 21874
· x7 − 15309 · x5 + 85050 · x3 − 105840 · x · cos 3x


+ −x10 + 63 4
· x6 − 1575 · x4 + 39690 · x2 − 136080 · sin


+ − 43 · x7 + 189 · x5 − 9450 · x3 + 105840 · x · cos x .

Then, we consider the previous expression as two separate expressions, the first with
positive and the second with negative terms next to sine and cosine functions



f + (x) = 42525 · x4 + 45360 · sin 3x + 189 · x5 + 105840 · x · cos x



+ 634
· x6 + 39690 · x2 · sin x + 2187
4
· x7 + 85050 · x3 · cos 3x ,



f − (x) = −x10 − 1575 · x4 − 136080 · sin x + −15309 · x5 − 105840 · x · cos 3x



+ − 15309
4
· x6 − 119070 · x2 · sin 3x + − 43 · x7 − 9450 · x3 · cos x .

III (Determination of downward rational polynomial approximation) After substitu-

tion of sine and cosine functions by appropriate (downward or upward) polynomial
approximations, we obtain downward polynomial approximations of f + (x) and f − (x)
respectively

sin,0
Pi+0 ,i1 ,i2 ,i3 (x) = 42525 · x4 + 45360 · T 4·i 0 +3
(3x)
5

cos,0
+ 189 · x + 105840 · x · T 4·i1 +2 (x)

sin,0
+ 634
· x6 + 39690 · x2 · T 4·i 2 +3
(x)
2187 7 3

cos,0
+ 4 · x + 85050 · x · T 4·i3 +2 (3x) ,

sin,0
Pi−4 ,i5 ,i6 ,i7 (x) = −x10 − 1575 · x4 − 136080 · T 4·i4 +1 (x)

cos,0
+ −15309 · x5 − 105840 · x · T 4·i5 (3x)

sin,0
+ − 15309
4
· x6 − 119070 · x2 · T 4·i6 +1 (3x)

cos,0
+ − 43 · x7 − 9450 · x3 · T 4·i7 (x) .

For concrete indices (i0 , i1 , i2 , i3 , i4 , i5 , i6 , i7 ) = (3, 3, 3, 3, 3, 3, 3, 3), we obtain

+
P3,3,3,3 (x) = 42525 · x4 + 45360 · T sin,0
4·3+3 (3x)


+ 189 · x + 105840 · x · T cos,0
5
4·3+2 (x)
63 6 2

sin,0
+ 4 · x + 39690 · x · T 4·3+3 (x)


+ 2187
4
· x7 + 85050 · x3 · T cos,0
4·3+2 (3x) ,

+
P3,3,3,3 335267731
(x) = − 11176704000 · x21 + 38742049
276756480
· x19 − 602542
225225
· x17
− 169795501
28828800
· x15 + 7838771
20592
·x 13
− 15677779
5280
· x11
+76650 · x7 − 165312 · x5 − 132300 · x3 + 241920 · x

32
and

sin,0
−
P3,3,3,3 (x) = −x10 − 1575 · x4 − 136080 · T 4·3+1 (x)

cos,0
−15309 · x5 − 105840 · x · T 4·3 (3x)

sin,0
+ − 15309
4
· x6 − 119070 · x2 · T 4·3+1 (3x)

cos,0
+ − 43 · x7 − 9450 · x3 · T 4·3 (x) ,
−
P3,3,3,3 1
(x) = − 6227020800 · x23 + 1
39916800
· x21 − 1355975527
1383782400
· x19
4289
+ 21621600 · x17 + 317827231
28828800
· x15 − 8632499
20592
·x 13
+ 5302897
1760
· x11
−76650 · x7 + 165312 · x5 + 132300 · x3 − 241920 · x .

Finally, for the MTP function

f (x) = f + (x) + f − (x)

we obtain the concrete downward polynomial approximation

+ −
P (x) = P3,3,3,3 (x) + P3,3,3,3 (x)
1
= − 6227020800 · x23 − 12417313
413952000
· x21 − 581132641
691891200
· x19
− 4449211
1663200
· x17 + 21361
4160
· x15 − 424
11
· x13 + 656
15
· x11

over (0, 1], i.e. it holds that

f (x) > P (x)
over (0, 1].
IV (The final part) Based on the Sturm theorem, the following inequality

P (x) > 0

is true over (0, 1]. The stated conclusion for the polynomial function P is correct based
on the following facts:
1. We can conclude, by Sturm theorem, that the polynomial function P (x) has only
one zero over the concrete extended segment [−0.1, 1] of the initial interval (0, 1].
2. Facts P (−0.1) 6= 0, P (0) = 0 and P (1) 6= 0 are correct.

3. The polynomial P is positive at boundary point 1 P (1) = 6.77772 . . . > 0 .
Therefore, the following inequality

f (x) > 0

is true over (0, 1]. 

Appendix A3
The initial MTP function is


h(x) = −52488 · x7 + 816480 · x5 − 3810240 · x3 + 4354560 · x · cos3 x


+ −6561·x8+244944·x6−2041200·x4+5080320·x2−1814400 ·sin x·cos2 x


+ −x11 + 39384 · x7 − 614880 · x5 + 2963520 · x3 − 4354560 · x · cos x


+ 1641 · x8 − 61488 · x6 + 529200 · x4 − 1693440 · x2 + 1814400 · sin x

33
  
and the initial interval is S = 1, π2 .

After the multiplication by −1, we obtain the MTP function

f (x) = 52488 · x7 − 816480 · x5 + 3810240 · x3 − 4354560 · x · cos3 x


+ 6561·x8−244944·x6+2041200·x4−5080320·x2+1814400 ·sin x·cos2 x


+ x11 − 39384 · x7 + 614880 · x5 − 2963520 · x3 + 4354560 · x · cos x


+ −1641 · x8 + 61488 · x6 − 529200 · x4 + 1693440 · x2 − 1814400 · sin x .

 
Automated proof that f (x) > 0 for x ∈ 1, π2 :

I (Recognition of possible case) Facts f (1) 6= 0 and f π2 6= 0 are correct.

The MTP
function f is positive

at boundary point 1

f (1) = 27.02986 . . . > 0 and at boundary
π
 π  π/2 f 2 = 5021.73462 . . . > 0 . Therefore, it is possible that f (x) > 0 over
point
1, 2 .

II (Transformation of angles) After the transformation of terms cosm x · sinn x (m, n ∈

N0 ) into the sum of sine and cosine functions of multiple angles, in the MTP function
f , we obtain

f (x) = 6561
4
· x8 − 61236 · x6 + 510300 · x4 − 1270080 · x2 + 453600 · sin 3x


+ 13122 · x7 − 204120 · x5 + 952560 · x3 − 1088640 · x · cos 3x


+ − 43 · x8 + 252 · x6 − 18900 · x4 + 423360 · x2 − 1360800 · sin x


+ x11 − 18 · x7 + 2520 · x5 − 105840 · x3 + 1088640 · x · cos x .

Then, we consider the previous expression as two separate expressions, the first with
positive and the second with negative terms next to sine and cosine functions

f + (x) = · x8 + 510300 · x4 + 453600 · sin 3x
6561
4


+ 13122 · x7 + 952560 · x3 · cos 3x


+ 252 · x6 + 423360 · x2 · sin x


+ x11 + 2520 · x5 + 1088640 · x · cos x ,


f − (x) = −61236 · x6 − 1270080 · x2 · sin 3x


+ −204120 · x5 − 1088640 · x · cos 3x


+ − 43 · x8 − 18900 · x4 − 1360800 · sin x


+ −18 · x7 − 105840 · x3 · cos x .

III (Determination of downward rational polynomial approximation) After substitu-

tion of sine and cosine functions by appropriate (downward or upward) polynomial
approximations, we obtain downward polynomial approximations of f + (x) and f − (x)

34
respectively

sin,0
Pi+0 ,i1 ,i2 ,i3 (x) = 4
· x8 + 510300 · x4 + 453600 · T 4·i
6561
0 +3
(3x)
7 3

cos,0
+ 13122 · x + 952560 · x · T 4·i1 +2 (3x)

sin,0
+ 252 · x6 + 423360 · x2 · T 4·i2 +3
(x)


+ x11 + 2520 · x5 + 1088640 · x · T cos,0
4·i3 +2 (x) ,

sin,0
Pi−4 ,i5 ,i6 ,i7 (x) = −61236 · x6 − 1270080 · x2 · T 4·i4 +1 (3x)

cos,0
+ −204120 · x5 − 1088640 · x · T 4·i5 (3x)

sin,0
+ − 43 · x8 − 18900 · x4 − 1360800 · T 4·i6 +1 (x)

cos,0
+ −18 · x7 − 105840 · x3 · T 4·i7 (x) .

For concrete indices (i0 , i1 , i2 , i3 , i4 , i5 , i6 , i7 ) = (3, 4, 1, 2, 4, 4, 2, 2), we obtain

+

sin,0
P3,4,1,2 (x) = · x8 + 510300 · x4 + 453600 · T 4·3+3
6561
4
(3x)

cos,0
+ 13122 · x7 + 952560 · x3 · T 4·4+2 (3x)


+ 252 · x6 + 423360 · x2 · T sin,0
4·1+3 (x)


+ x + 2520 · x + 1088640 · x · T cos,0
11 5
4·2+2 (x) ,

+
P3,4,1,2 387420489
(x) = − 487911424000 · x25 + 129140163
14350336000
· x23 − 6175128960
2208298489
· x21
+ 129141043
35481600
· x19 − 760514951
16473600
· x17 + 607555181
2882880
· x15
+ 53310093
22880
· x13 − 13876823
440
· x11 + 39372 · x9 + 735840 · x7
−1859760 · x5 − 1209600 · x3 + 2449440 · x

and

sin,0
−
P4,4,2,2 (x) = −61236 · x6 − 1270080 · x2 · T 4·4+1 (3x)

cos,0
+ −204120 · x5 − 1088640 · x · T 4·4 (3x)

sin,0
+ − 43 · x8 − 18900 · x4 − 1360800 · T 4·2+1 (x)

cos,0
+ −18 · x7 − 105840 · x3 · T 4·2 (x) ,
−
P4,4,2,2 387420489
(x) = − 17425408000 · x23 + 129140163
512512000
· x21 − 43046721
8712704
· x19
+ 19715397503
345945600
· x17 − 127545983
480480
· x15 − 531449
240
· x13
+ 314933
10
· x11 − 39372 · x − 735840 · x
9 7

+1859760 · x5 + 1209600 · x3 − 2449440 · x .

Finally, for the MTP function

f (x) = f + (x) + f − (x)

35
we obtain the concrete downward polynomial approximation
+ −
P (x) = P3,4,1,2 (x) + P4,4,2,2 (x)
387420489
= − 487911424000 · x25 − 129140163
9758228480
· x23 − 521856760043
4940103168000
· x21
− 10201878397
7841433600
· x19 + 936145883
86486400
· x17 − 2048321
37440
· x15
+ 1587173
13728
· x13 − 19771
440
· x11
 
over 1, π2 , i.e. it holds that
f (x) > P (x)
 π

over 1, 2
.
IV (The final part) Based on the Sturm theorem, the following inequality

P (x) > 0
 π

is true over 1, 2 . The stated conclusion for the polynomial function P is correct
based on the following facts:
1. We can conclude, by Sturm theorem, that the polynomial function P (x) doesnot
have zero over the concrete extended segment [1, 1.58] of the initial interval 1, π2 .
2. Facts P (1) 6= 0 and P (1.58) 6= 0 are correct.


3. The polynomial P is positive at boundary point π2 P π2 = 1228.02881 . . . > 0 .
Therefore, the following inequality

f (x) > 0
 π

is true over 1, 2
. 

Appendix A4
The initial MTP function is


f (x) = x5 − 360 · x + −30 · x2 + 630 · sin x − 270 · x · cos x


and the initial interval is S = 0, π2 .


Automated proof that f (x) > 0 for x ∈ 0, π2 :
π


I (Recognition of possible case) Facts f (0) = 0 and
f 2 6= 0 are correct.

The MTP
π π
function f is positive at boundary
point 2 f 2 = 0.054404 . . . > 0 . Therefore, it
is possible that f (x) > 0 over 0, π2 .
II (Transformation of angles) After the transformation of terms cosm x · sinn x (m, n ∈
N0 ) into the sum of sine and cosine functions of multiple angles, in the MTP function
f , we obtain


f (x) = −30 · x2 + 630 · sin x − 270 · x · cos x + x5 − 360 · x .

Then, we consider the previous expression as two separate expressions, the first with
positive and the second with negative terms next to sine and cosine functions

f + (x) = 630 · sin x + x5 ,

f − (x) = −360 · x − 270 · x · cos x − 30 · x2 · sin x .

36
III (Determination of downward rational polynomial approximation) After substitu-
tion of sine and cosine functions by appropriate (downward or upward) polynomial
approximations, we obtain downward polynomial approximations of f + (x) and f − (x)
respectively
sin,0
Pi+0 (x) = 630 · T 4·i0 +3
(x) + x5 ,
cos,0 sin,0
Pi−1 ,i2 (x) = −360 · x − 270 · x · T 4·i1 (x) − 30 · x2 · T 4·i2 +1 (x) .
For concrete indices (i0 , i1 , i2 ) = (2, 2, 2), we obtain
P2+ (x) = 630 · T sin,0 5
4·2+3 (x) + x ,

P2+ (x) = − 63360

1
· x11 + 1
576
· x9 − 1
8
· x7 + 25
4
· x5 − 105 · x3 + 630 · x
and
cos,0 sin,0
−
P2,2 (x) = −360 · x − 270 · x · T 4·2 (x) − 30 · x2 · T 4·2+1 (x) ,
−
P2,2 1
(x) = − 12096 · x11 − 1
1344
· x9 + 1
8
· x7 − 25
4
· x5 + 105 · x3 − 630 · x .
Finally, for the MTP function
f (x) = f + (x) + f − (x)
we obtain the concrete downward polynomial approximation
P (x) = P2+ (x) + P2,2
−
(x)
131
= − 1330560 · x11 + 1
1008
· x9
π


over 0, 2
, i.e. it holds that
f (x) > P (x)
π


over 0, 2
.
IV (The final part) Based on the Sturm theorem, the following inequality
P (x) > 0

π
is true over 0, . The stated conclusion for the polynomial function P is correct
2
based on the following facts:
1. We can conclude, by Sturm theorem, that the polynomial function P (x) has only

one zero over the concrete extended segment [−0.1, 1.58] of the initial interval 0, π2 .
2. Facts P (−0.1) 6= 0, P (0) = 0 and P (1.58) 6= 0 are correct.



3. The polynomial P is positive at boundary point π2 P π2 = 0.043615 . . . > 0 .
Therefore, the following inequality
f (x) > 0

π
is true over 0, . 2


Appendix A5
The initial MTP function is


f (x) = x8 − 21 · x6 + 630 · x4 − 7560 · x2 + 15120 · sin x


+ 3 · x7 − 126 · x5 + 2520 · x3 − 15120 · x · cos x


and the initial interval is S = 0, π2 .

37
 π


Automated proof that f (x) > 0 for x ∈ 0, 2
:


I (Recognition of possible case) Facts f (0) = 0 and

f π2 6= 0 are correct.

The MTP
π π
function f is positive at boundary point

2
f 2
= 23.53938 . . . > 0 . Therefore, it
is possible that f (x) > 0 over 0, π2 .
II (Transformation of angles) After the transformation of terms cosm x · sinn x (m, n ∈
N0 ) into the sum of sine and cosine functions of multiple angles, in the MTP function
f , we obtain


f (x) = x8 − 21 · x6 + 630 · x4 − 7560 · x2 + 15120 · sin x


+ 3 · x7 − 126 · x5 + 2520 · x3 − 15120 · x · cos x .

Then, we consider the previous expression as two separate expressions, the first with
positive and the second with negative terms next to sine and cosine functions



f + (x) = x8 + 630 · x4 + 15120 · sin x + 3 · x7 + 2520 · x3 · cos x ,



f − (x) = −126 · x5 − 15120 · x · cos x + −21 · x6 − 7560 · x2 · sin x .

III (Determination of downward rational polynomial approximation) After substitu-

tion of sine and cosine functions by appropriate (downward or upward) polynomial
approximations, we obtain downward polynomial approximations of f + (x) and f − (x)
respectively

sin,0


Pi+0 ,i1 (x) = x8 + 630 · x4 + 15120 · T 4·i0 +3
(x) + 3 · x7 + 2520 · x3 · T cos,0
4·i1 +2 (x) ,

cos,0
sin,0
Pi−2 ,i3 (x) = −126 · x5 − 15120 · x · T 4·i2 (x) + −21 · x6 − 7560 · x2 · T 4·i3 +1 (x) .

For concrete indices (i0 , i1 , i2 , i3 ) = (1, 1, 2, 2), we obtain

sin,0


+
P1,1 (x) = x8 + 630 · x4 + 15120 · T 4·1+3 (x) + 3 · x7 + 2520 · x3 · T cos,0
4·1+2 (x) ,

+
P1,1 1
(x) = − 5040 · x15 + 1
240
· x13 − 1
6
· x11 + 5
4
· x9 − 504 · x5 + 15120 · x

and

cos,0
sin,0
−
P2,2 (x) = −126 · x5 − 15120 · x · T 4·2 (x) + −21 · x6 − 7560 · x2 · T 4·2+1 (x) ,
−
P2,2 1
(x) = − 17280 · x15 + 1
960
· x13 − 1
48
· x11 − 5
8
· x9 + 504 · x5 − 15120 · x .

Finally, for the MTP function

f (x) = f + (x) + f − (x)

we obtain the concrete downward polynomial approximation

+ −
P (x) = P1,1 (x) + P2,2 (x)
31
= − 120960 · x15 + 1
192
· x13 − 3
16
· x11 + 5
8
· x9
π


over 0, 2
, i.e. it holds that
f (x) > P (x)

π
over 0, . 2

38
IV (The final part) Based on the Sturm theorem, the following inequality

P (x) > 0


is true over 0, π2 . The stated conclusion for the polynomial function P is correct
based on the following facts:
1. We can conclude, by Sturm theorem, that the polynomial function P (x) has only

one zero over the concrete extended segment [−0.1, 1.58] of the initial interval 0, π2 .
2. Facts P (−0.1) 6= 0, P (0) = 0 and P (1.58) 6= 0 are correct.



3. The polynomial P is positive at boundary point π2 P π2 = 11.074847 . . . > 0 .
Therefore, the following inequality

f (x) > 0

is true over (0, π/2). 

Appendix A6
The initial MTP function is


f (x) = 3 · x5 − 20 · x − 140 · sin x · cos x + 30 · x2 − 70 · sin x
+40 · x · cos2 (x) + 190 · x · cos x
π


and the initial interval is S = 0, 2
.
π


Automated proof that f (x) > 0 for x ∈ 0, 2
:


I (Recognition of possible case) Facts f (0) = 0 and

f π2 6= 0 are correct.

The MTP
π π
function f is positive at boundary
point 2
f 2
= 1.29545 . . . > 0 . Therefore, it is
possible that f (x) > 0 over 0, π2 .
II (Transformation of angles) After the transformation of terms cosm x · sinn x (m, n ∈
N0 ) into the sum of sine and cosine functions of multiple angles, in the MTP function
f , we obtain


f (x) = −70 · sin 2x + 20 · x · cos 2x + 30 · x2 − 70 · sin x + 190 · x · cos x + 3 · x5 .

Then, we consider the previous expression as two separate expressions, the first with
positive and the second with negative terms next to sine and cosine functions

f + (x) = 190 · x · cos x + 20 · x · cos 2x + 30 · x2 · sin x + 3 · x5 ,

f − (x) = −70 · sin x − 70 · sin 2x .

III (Determination of downward rational polynomial approximation) After substitu-

tion of sine and cosine functions by appropriate (downward or upward) polynomial
approximations, we obtain downward polynomial approximations of f + (x) and f − (x)
respectively

Pi+0 ,i1 ,i2 (x) = 190 · x · T cos,0 cos,0 2 sin,0 5

4·i0 +2 (x) + 20 · x · T 4·i1 +2 (2x) + 30 · x · T 4·i2 +3 (x) + 3 · x ,

sin,0 sin,0
Pi−3 ,i4 (x) = −70 · T 4·i3 +1 (x) − 70 · T 4·i4 +1 (2x) .

39
For concrete indices (i0 , i1 , i2 , i3 , i4 ) = (1, 2, 1, 2, 2), we obtain
+
P1,2,1 (x) = 190 · x · T cos,0 cos,0 2 sin,0 5
4·1+2 (x) + 20 · x · T 4·2+2 (2 · x) + 30 · x · T 4·1+3 (x) + 3 · x ,

+
P1,2,1 16
(x) = − 2835 · x11 + 61
504
· x9 − 43
24
· x7 + 77
4
· x5 − 105 · x3 + 210 · x

and
− sin,0 sin,0
P2,2 (x) = −70 · T 4·2+1 (x) − 70 · T 4·2+1 (2x) ,
−
P2,2 19
(x) = − 192 · x9 + 43
24
· x7 − 77
4
· x5 + 105 · x3 − 210 · x .
Finally, for the MTP function

f (x) = f + (x) + f − (x)

we obtain the concrete downward polynomial approximation

+ −
P (x) = P1,2,1 (x) + P2,2 (x)
16
= − 2835 · x11 + 89
4032
· x9
π


over 0, 2
, i.e. it holds that
f (x) > P (x)

π
over 0, . 2

IV (The final part) Based on the Sturm theorem, the following inequality

P (x) > 0


is true over 0, π2 . The stated conclusion for the polynomial function P is correct
based on the following facts:
1. We can conclude, by Sturm theorem, that the polynomial function P (x) has only

one zero over the concrete extended segment [−0.1, 1.58] of the initial interval 0, π2 .
2. Facts P (−0.1) 6= 0, P (0) = 0 and P (1.58) 6= 0 are correct.



3. The polynomial P is positive at boundary point π2 P π2 = 0.47438 . . . > 0 .
Therefore, the following inequality

f (x) > 0


is true over 0, π2 . 

Appendix A7
The initial MTP function is


f (x) = = 16 · x6 + 135 · x · sin x · cos2 (x) + 90 · x3 − 360 · x · sin x


+360 · cos3 (x) + 585 · x2 − 360 · cos x


and the initial interval is S = 0, π2 .


Automated proof that f (x) > 0 for x ∈ 0, π2 :


I (Recognition of possible case) Facts f (0) = 0 and
f π2 6= 0 are correct.

The MTP
π π
function f is positive at boundary point

2
f 2
= 23.68123 . . . > 0 . Therefore, it
is possible that f (x) > 0 over 0, π2 .

40
II (Transformation of angles) After the transformation of terms cosm x · sinn x (m, n ∈
N0 ) into the sum of sine and cosine functions of multiple angles, in the MTP function
f , we obtain



f (x) = = 90 · cos 3 · x + 90 · x3 − 1305
4
· x · sin x + 585 · x2 − 90 · cos x
+16 · x6 + 135
4
· x · sin 3x .

Then, we consider the previous expression as two separate expressions, the first with
positive and the second with negative terms next to sine and cosine functions

f + (x) = 90 · cos 3x + 135

4
· x · sin 3x + 585 · x2 · cos x + 90 · x3 · sin x + 16 · x6 ,
1305
f − (x) = −90 · cos x − 4
· x · sin x .

III (Determination of downward rational polynomial approximation) After substitu-

tion of sine and cosine functions by appropriate (downward or upward) polynomial
approximations, we obtain downward polynomial approximations of f + (x) and f − (x)
respectively

Pi+0 ,i1 ,i2 ,i3 (x) = 90 · T cos,0

4·i0 +2 (3x) +
135
4
sin,0
· x · T 4·i1 +3
(3x)
+585 · x2 · T cos,0 3 sin,0 6
4·i2 +2 (x) + 90 · x · T 4·i3 +3 (x) + 16 · x ,

cos,0 1305 sin,0

Pi−4 ,i5 (x) = −90 · T 4·i4 (x) − 4
· x · T 4·i5 +1 (x) .

For concrete indices (i0 , i1 , i2 , i3 , i4 , i5 ) = (2, 3, 1, 1, 2, 2), we obtain

+
P2,3,1,1 (x) = 90 · T cos,0
4·2+2 (3x) +
135
4
· x · T sin,0
4·3+3 (3x)

+585 · x2 · T cos,0 3 sin,0 6

4·1+2 (x) + 90 · x · T 4·1+3 (x) + 16 · x ,

+
P2,3,1,1 531441
(x) = − 1435033600 · x16 + 177147
20500480
· x14 − 59049
394240
· x12 + 6241
17920
· x10
1
− 16 · x8 + 83
32
· x6 − 405
8
· x4 + 1125
4
· x2 + 90

and
− cos,0 1305 sin,0
P2,2 (x) = −90 · T 4·2 (x) − 4
· x · T 4·2+1 (x) ,
−
P2,2 29
(x) = − 32256 · x10 + 1
16
· x8 − 83
32
· x6 + 405
8
· x4 − 1125
4
· x2 − 90 .

Finally, for the MTP function

f (x) = f + (x) + f − (x)

we obtain the concrete downward polynomial approximation

+ −
P (x) = P2,3,1,1 (x) + P2,2 (x)
531441
= − 1435033600 · x16 + 177147
20500480
· x14 − 59049
394240
· x12 + 7003
20160
· x10
π


over 0, 2
, i.e. it holds that
f (x) > P (x)

π
over 0, . 2

41
IV (The final part) Based on the Sturm theorem, the following inequality
P (x) > 0
π


is true over 0, . The stated conclusion for the polynomial function P is correct
2
based on the following facts:
1. We can conclude, by Sturm theorem, that the polynomial function P (x) has only

one zero over the concrete extended segment [−0.1, 1.58] of the initial interval 0, π2 .
2. Facts P (−0.1) 6= 0, P (0) = 0 and P (1.58) 6= 0 are correct.



3. The polynomial P is positive at boundary point π2 P π2 = 2.27261 . . . > 0 .
Therefore, the following inequality
f (x) > 0


is true over 0, π2 . 

Bojan Banjac
University of Novi Sad,
Faculty of Technical Sciences,
Trg Dositeja Obradovića 16,
21000 Novi Sad, Serbia,
E-mail: bojan.banjac@uns.ac.rs

Branko Malešević
University of Belgrade,
School of Electrical Engineering,
Bulevar kralja Aleksandra 73,
11000 Belgrade, Serbia,
E-mail: branko.malesevic@etf.bg.ac.rs

Miloš Mićović
University of Belgrade,
School of Electrical Engineering,
Bulevar kralja Aleksandra 73,
11000 Belgrade, Serbia,
E-mail: milos.micovic@etf.bg.ac.rs

Bojana Mihailović
University of Belgrade,
School of Electrical Engineering,
Bulevar kralja Aleksandra 73,
11000 Belgrade, Serbia,
E-mail: mihailovicb@etf.bg.ac.rs

Milica Savatović
University of Belgrade,
School of Electrical Engineering,
Bulevar kralja Aleksandra 73,
11000 Belgrade, Serbia,
E-mail: milica.makragic@etf.bg.ac.rs

42