Some tight polynomial-exponential lower bounds for an exponential function

Transcription

1 Some tight polynomial-exponential lower bounds for an exponential function Christophe Chesneau To cite this version: Christophe Chesneau. Some tight polynomial-exponential lower bounds for an exponential function. 18. <hal-16864> HAL Id: hal Submitted on 1 Jan 18 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

2 Some tight polynomial-exponential lower bounds for an exponential function Christophe Chesneau a,1 a LMNO, University of Caen Normandie, France Abstract This note is devoted to new sharp lower bounds for exp(x. We first introduce and study a new lower bound defined with polynomial of degree and exponential (or hyperbolic functions. Then we propose two improvements of this lower bound by using two different approaches; the first approach consists in adding well-chosen polynomial term to it, whereas the second approach aims to transform it for large values of x. We show that they are better to well-known lower bounds. The analytic results are supported by some numerical studies and graphics. A part of the study is devoted to some integral methods having the ability to generate new lower bounds for exp(x. Keywords: Algebraic bounds, exponential function. MSC: 33B1, 6D7. 1. Introduction Inequalities involving exponential functions are useful in all the areas of mathematics. The most famous of them can be found in Mitrinović (1964, Bullen (1998 and Kuang (1993. See also Qi (1997, Bae (7, Kim (7 and Bae and Kim (9 for current developments on lower and upper bound for exp(x. The purpose of this note is to provide simple and tight lower bounds for exp(x. Such bounds are important tools to evaluate lower or upper bounds of mathematical terms involving exp(x. Basic examples include the functions cosh(x, sinh(x and tanh(x, the integral x exp(y dy, the sum exp( k, the cumulative distribution function of the Gaussian or Kolmogorov distributions. Well-known lower bounds for exp(x are cosh( x, sinh ( 6x /( 6x, exp(x x, 1 + x + x 4 / and ( 1 + x /a a with a >. The last one is sharp for x large only for large values of a, corresponding to a polynomial with a high degree and large coefficients when a is an integer. Recent sharp lower bounds can be found in Bagul (17 for x is a an interval of the form [, b] with a precise value for b. The motivation of this paper is to introduce new sharp lower bounds for exp(x defined with simple functions, at least uniformly better to the two benchmarks: cosh( x and 1 + x + x 4 /, for all x R. In a first part, a first lower bound is introduced. It is defined as an even function on R with simple polynomial of degree and exponential functions (without power of x. Comparison to cosh( x, exp(x x and 1 + x + x 4 / are made analytically and with the used of graphics. Then we propose two significant improvements of this lower bound via two different approaches. The first approach aims to add well chosen polynomial terms to the former lower bound. The second approach adopts the transformation suggested in Chesneau (17. It consists in weighting and translating the former lower bound when x is large enough. Only polynomial of degree and exponential functions are used. In each case, the theoretical results are supported by a short numerical study and some graphics, illustrating the tightness of the new lower bounds. k= 1 christophe.chesneau@unicaen.fr 1

3 Finally, we present new integration approaches to generate lower bounds for exp(x using existing lower bounds. The link existing with this approach and the main lower bound of the study is discussed. The rest of this note is structured as follows. Section presents our main lower bound. Section 3 is devoted to an improvement of this lower bound. Another improvement is developed in Section 4. Section 5 presents some integral approaches to determine lower bounds for exp(x. All the proofs of our results are postponed in Section 6.. Lower bound The main lower bound for exp(x is presented in the proposition below. Proposition 1. Let us define the function f(x by f(x = 1 [ ( exp x (7 4 ( x + x + exp x (7 + 4 ] x + x 1. (1 Then, for all x R, exp(x f(x 1. Let us observe that f(x is a continuous even function on R using simple polynomial-exponential functions. It is of the form f(x = (1/(G(x + G( x 1, where G(x = exp ( x (7 4 x + x. It can be also express in terms of hyperbolic cosh and sinh functions as: ( f(x = (7 + x cosh x 4 ( x sinh x 6. ( A visual comparison between exp(x and f(x is performed in Figure 1, for x [ 1, 1] for the first graphic and x [.5,.5] for the second graphic. We can see that the two curves are relatively close, specially for small value for x. This comment is also supported by Table 1 which indicates the numerical values of the error measure: R(b = b [ b f(x exp(x ] dx, for several values for b. The numerical studies are done with the software Mathematica (version 11, see Wolfram ( expx 6 expx fx 4 fx Figure 1: Superimposed curves of exp(x and f(x for x [ 1, 1] in the first graphic, and for x [.5,.5] in the second graphic. Table 1: Numerical evaluations of R(b for b {.5, 1, 1.5,,.5, 3}. b =.5 b = 1 b = 1.5 b = b =.5 b = 3 R(b The tightness of f(x is highlighted in Proposition below; we proves that f(x is uniformly greater to most well-known simple lower bounds for exp(x : cosh ( x, exp(x x, 1 + x + x 4 /.

4 Proposition. Let f(x be the function given by (1. We have, for all x R, [ ( ] f(x max cosh x, exp(x x, 1 + x + x4. Hence f(x can be viewed as the best lower bounds among cosh ( x, exp(x x and 1 + x + x 4 /. Figure illustrates this result by considering two intervals of values for x : [ 1, 1] and [ 3, 3], one for each graphics. It shows that f(x is closer to exp(x is comparison to cosh ( x, exp(x x and 1 + x + x 4 /..5 5 f(x 4 f(x. 1+x + x4 3 1+x + x4 cosh x cosh x 1.5 exp(x-x exp(x-x ( x Figure : Superimposed curves of f(x, 1 + x + x 4 /, cosh and exp(x x for x [ 1, 1] in the first graphic, and for x [ 3, 3] in the second graphic. Since cosh ( x 1 or exp(x x 1 or 1 + x + x 4 / 1, Proposition implies that f(x 1, which is the second inequality in Proposition 1. If we consider the polynomial of degree 6: 1 + x + x 4 / + x 6 /6, which is also a loser bound for exp(x, we have f(x 1 + x + x 4 / + x 6 /6 for some x, but their exists x such that the reverse holds. This motivates the study of some improvements of f(x, which concerns the rest of the study. Two modifications are proposed: adding a well-chosen polynomial term to f(x or transforming f(x for x large. Remark 1. Let us mention that some continuous even upper bounds for exp(x using hyperbolic cosh and sinh functions can be proved. An example with x bounded is the following: For any a > and x a, we have exp(x cosh (ax + (x/a sinh (ax ( exp(a x. 3. First improvement of the lower bound First of all, let us consider an intermediary result which can be viewed as an improvement of the wellknown inequality: for all x R, exp(x cosh ( x. Lemma 1. For all x R and n N, we define the polynomial P n (x by Then, for all x R and n N, we have P n (x = x 4 n k= exp(x cosh x k k!(k + 3(k + ( x + P n (x. Since P n (x for all x R and n N, it is clear that exp(x cosh ( x + P n (x cosh ( x. Let us observe that P n (x satisfies the recurrence relation: P n (x = P n 1 (x + x (n+ /(n!(n + 3(n +, with P (x = x 4 /3. Expressions of P n (x for several values of n are given in Table. Lemma 1 is a key tool to the proof of the following proposition in which we determine a tight lower bound of exp(x uniformly greater to f(x. 3

5 Table : Analytic expressions for P n(x with n {,..., 5}. n = n = 1 n = n = 3 P n (x 1 3 x4 1 3 x x6 1 3 x x x8 1 3 x x x x1 n = 4 n = x x x x x1 1 3 x x x x x x14 Proposition 3. For all x R and n N, we define the polynomial Q n (x by n Q n (x = 4x 6 x k ( k!(k + 3(k + k= 1 (k + 5(k x (k + 4(k + 7 Let f(x be the function given by (1. We define the function f (x; n by Then, for all x R and n N, we have f (x; n = f(x + Q n (x. exp(x f (x; n f(x. Thus f (x; n is a better lower to f(x for exp(x, for all x R. Remark that Q n (x satisfies the recurrence relation: Q n (x = Q n 1 (x+4x (n+3 /(n!(n+3(n+[1/((n+5(n+3+x /((n+4(n+7], with Q (x = x 6 /45 + x 8 /4. Expressions of Q n (x for several values of n are given in Table 3.. Table 3: Analytic expressions for Q n(x with n {,..., 5}. n = n = 1 n = Q n (x 1 45 x x x x x x x x x1 n = 3 n = x x x x x x x x x x x16 The result of Proposition 3 is illustrated in Figure 3. The two graphics consider the two intervals for x: [ 1, 1] and [.5,.5] expx f (x, f(x 6 4 expx f (x, f(x Figure 3: Superimposed curves of exp(x, f (x; n for n = and f(x for x [ 1, 1] in the first graphic, and for x [.5,.5] in the second graphic. Table 4 shows the numerical values of the error measure: R (b = b [ f (x; n exp(x ] b dx, for n = and several values for b. 4

6 Table 4: Numerical evaluations of R (b for b {.5, 1, 1.5,,.5, 3}. b =.5 b = 1 b = 1.5 b = b =.5 b = 3 R (b Second improvement of the lower bound We now investigate a transformation of f(x for large x, based on a multiplicative exponential weight and translation. It is an adaptation of the method developed by Chesneau (17 to f(x. Proposition 4. Let f(x be the function given by (1. For any a, we define the function f(x; a by f (x; a = f(x1 { x <a/} (x + f( x a exp(a x a 1 { x a/} (x, where 1 A (x denotes the indicator function over A, i.e. 1 A (x = 1 if x A and elsewhere. Then, for all a and x R, we have exp(x f (x; a f(x. Another look of the function f (x; a is given by f (x; a = max [ f(x, f( x a exp(a x a ]. Note that f (x; a is an even continuous function on R. It follows from Proposition 4 that f (x; a is a better lower to f(x for exp(x, for all a and x R. Figure 4 proposes a graphical illustration of Proposition 4. The two graphics consider the two intervals respectively: [ 1, 1] and [ 3, 3]. We see that exp(x and f (x, a with a = 1 are near confounded for the considered values for x, showing the sharpness of the lower bound. Owing to Proposition, we have f (x; a max [ ( cosh x, 1 + x + x4 [ ( + max cosh ( x a, ] 1 { x <a/} (x (1 + ( x a + ] ( x a4 exp(a x a 1 { x a/} (x. Therefore, defining with the same a, this lower bound is sharper to the lower bounds exhibited in Chesneau ( expx fox,1 fx 3 expx fox,1 fx Figure 4: Superimposed curves of exp(x, f (x; a for a = 1 and f(x for x [ 1, 1] in the first graphic, and for x [.5,.5] in the second graphic. Table 5 shows the numerical values of the error measure: R (b = b [ f (x; a exp(x ] b dx, for a = 1 and several values for b. The value a = 1 in the numerical study is arbitrary chosen; one can find a more optimal value for the problem for a given criteria of optimization. 5

7 Table 5: Numerical evaluations of R (b for b {.5, 1, 1.5,,.5, 3}. b =.5 b = 1 b = 1.5 b = b =.5 b = 3 R (b A comparison of the two lower bounds f (x; n and f (x; a is now discussed. When x is small, say x < a/ for a fixed a, we have f (x; n f(x = f (x; a, so f (x; n is better. When x is large, in view of Figure 4, and Tables 4 and 5, we claim that f (x; n is better for some a and n. As a direct applications of our results, let us mention that sharp polynomial-exponential lower bounds for exp(x give sharp polynomial-exponential lower bounds for cosh(x or sinh(x ; using the inequality exp(y 1 + y, for all y R, for any ψ(x {f(x, f (x; n, f (x; a}, for all x R, we have cosh(x = exp(x + exp( x 1 ( ψ(x + max(1 x,. On the other hand, using the inequality: exp( y 1 y + y /, for all y, for any x R, we have sinh(x = exp(x exp( x 1 These inequalities can be usefull in various mathematical settings. ( ψ(x max (1 x + x4, On some generators of lower bounds We now present and discuss some general approaches based on integration to generate new lower bounds for exp(x from existing lower bounds. When it is possible, conditions are mentioned to improved the tightness of the former lower bounds First integral approach The main result is described in the proposition below. Proposition 5. Let θ(x be a positive function on R and ω(x be the function defined by ω(x = (1 + t θ(tdt dy + 1. Suppose that exp(x θ(x for all x R. Then exp(x ω(x for all x R. Suppose that θ(x is even, two times differentiable with θ( = 1, θ ( = and (1+x θ(x θ (x for all x. Then, for all x R, we have ω(x θ(x. Under the assumptions of the two points above, we have So ω(x is a better lower bound to θ(x. exp(x ω(x θ(x. 6

8 Connections between Proposition 5 and 1 exist. Indeed, let us consider the well-known lower bound for exp(x : θ(x = cosh ( x. We have θ (x = sinh ( x and θ (x = cosh ( x. Therefore θ( = 1, θ ( = and (1 + x θ(x θ (x = 4x cosh ( x. It follows from Proposition 5 that a better lower bound of θ(x = cosh ( x is given by ω(x = = (1 + t θ(tdt dy + 1 ( ( 4y cosh y = (7 + x cosh + 3 ( sinh y ( x 4 ( x sinh x 6. + ( y sinh y dy + 1 We thus obtain the hyperbolic expression of the lower bound f(x given by (. Naturally, the first point of Proposition 5 can be used to generate new lower bounds for exp(x. For instance, using the inequality exp(y 1 + y for all y R, we have exp(x = exp(x x exp( x (1 + x x exp( x. Let us set θ(x = (1 + x x exp( x. Hence a new lower bound for exp(x is given by ω(x = (1 + t θ(tdt dy + 1 = [ exp(y ( y + 66y y 3 + 4y ] dy + 1 = exp( x ( x + 174x 36 x 3 + 4x x. However, note that this lower bound is not better to θ(x. In particular, the assumption (1 + x θ(x θ (x for all x R is not satisfied. This bound is tight but not shaper than f(x for all x R. Moreover, from a mathematical point of view, it is more difficult to manipulate to f(x. Let us mention that the well-known lower bounds: 1 + x + x 4 / and exp( x x, also satisfy θ( = 1, θ ( = and (1 + x θ(x θ (x for all x, yielding more sharp lower bounds ω(x for exp(x. However, one can show that they are not better to f(x for all x R (and the presented improvements. 5.. Generalization Proposition 6 below presents a generalization of Proposition 5. From two lower bounds θ 1 (x and θ (x of exp(x, one can construct a lower bound better to θ 1 (x or θ (x, under some assumptions. Proposition 6. Let θ 1 (x and θ (x be two positive functions on R and κ(x be the function defined by κ(x = (θ 1 (t + t θ (tdt dy + 1. Suppose that exp(x max [θ 1 (x, θ (x] for all x R. Then exp(x κ(x for all x R. Suppose that θ 1 (x and θ (x are even, two times differentiable with θ 1 ( = 1, θ 1( =, (θ 1 (x + x θ (x θ 1 (x for all x. Then, for all x R, we have κ(x θ 1 (x. This inequality holds with θ (x by inverting the role of θ 1 (x and θ (x in the definition of κ(x and the conditions. θ 1 ( = 1, θ ( = 1, θ 1( =, θ ( =, (θ 1 (x + x θ (x θ 1 (x and (θ 1 (x + x θ (x θ (x for all x. Then, for all x R, we have κ(x max [θ 1 (x, θ (x]. 7

9 Under the assumptions of the first point and the second item of the second point, we have exp(x κ(x max [θ 1 (x, θ (x]. So κ(x is a better lower bound to θ 1 (x for exp(x, or both of them. Taking θ 1 (x = θ (x in Proposition 6, we obtain Proposition 5 with θ(x = θ 1 (x. Note that, taking θ 1 (x = cosh ( x and θ (x = n k= x k /(k!, the first point and the first item of the second point of Proposition 6 are satisfied; we thus obtain Lemma 1. Another example is given by chosing θ 1 (x = f(x and θ (x = cosh ( x. After some calculus, we have ( ( ( κ(x = 7 cosh x x 3 cosh x 1 ( x sinh x 6. Also, one can show that the first point and the second item of the second point of Proposition 6 are satisfied; κ(x is a better lower bound to f(x. However, the comparison with the proposed improvements for f(x need further investigations Another integral approach We conclude this section by presenting another generator of lower bounds for exp(x. Proposition 7. Let θ(x be a positive function on R and γ(x be the function defined by γ(x = exp( x exp(t [(1 + tθ(t 1] dt + 1. If exp(x θ(x for all x R. Then exp(x ω(x for all x R. For instance, if θ(x = cosh ( x, we have ( γ(x = 8 exp( x (7 + x cosh x + ( (5 + x sinh x. However, one can show that γ(x is not better to f(x for all x R. 6. Proofs Proof of Proposition 1. Let us set g(x = exp(x f(x. We aim to study this function. It follows from several algebraic manipulations that g (x = x exp(x 1 [ ( exp x (3 4x + ( x exp x (3 + 4x + ] x. (3 In order to study the sign of g (x, let us investigate g (x. Algebraic manipulations and simplifications give [ ( ] g (x = (1 + x exp(x cosh x. (4 Owing to the elementary inequality: exp(x cosh ( x, we have g (x. Thus g (x is increasing. Since g ( =, the minimum of g(x is attained in x = : for all x R, we have g(x g( =, implying that exp(x f(x. The second inequality can be prove in a similar manner. Let us set h(x = f(x 1. It follows from several algebraic manipulations that h (x = 1 [ ( exp x (3 4x + ( x exp x (3 + 4x + ] x 8

10 and ( h (x = (1 + x cosh x. So h (x and h (x is increasing. Since h ( =, for all x R, we have h(x h( =. Hence f(x 1. This ends the proof of Proposition 1. Proof of Proposition. Let us prove that, for all x R, f(x cosh ( x, f(x exp(x x and f(x 1 + x + x 4 /, in turn. Proof for f(x cosh ( x. Let us set k(x = f(x cosh ( x. After calculus and simplifications, we obtain k (x = ( x + ( x exp x ( + x + ( x exp x and ( k (x = 4 cosh x x. Since k (x, k (x is increasing. Since k ( =, for all x R, we have k(x k( =, implying the desired inequality. Proof for f(x exp(x x. Let us set l(x = f(x exp(x + x. After calculus and simplifications, we obtain l (x = 1 [ ( exp x (3 4x + ( x exp x (3 + 4x + ] x exp(x + and ( l (x = (1 + x cosh x exp(x. We have l (x cosh ( x exp(x, so l (x is increasing. Since l ( =, for all x R, we have l(x l( =, implying the desired inequality. Proof for f(x 1 + x + x 4 /. Let us set m(x = f(x (1 + x + x 4 /. After calculus and simplifications, we obtain m (x = 1 [ ( exp x (3 4x + ( x exp x (3 + 4x + ] x 4x(1 + x and ( m (x = (1 + x cosh x 6x. Since cosh ( x = + ( k x /(k! > 1+x, we have m (x (1+x (1+x 6x = 4x 4. k= Thus m (x is increasing. Since m ( =, for all x R, we have m(x m( =, implying the desired inequality. The proof of Proposition is complete. On Remark 1. Let us set ρ(x = (1/[1 + (x/a]. Since x a, observe that ρ(x [, 1]. Also, we can write x = ρ(x(ax + (1 ρ(x( ax. Owing to the convexity of the function exp(y, we have This completes Remark 1. exp(x = exp [ρ(x(ax + (1 ρ(x( ax] ρ(x exp(ax + (1 ρ(x exp( ax = cosh (ax + x sinh (ax. a 9

11 Proof of Lemma 1. Let us set p(x = exp(x cosh ( x P n (x. Then we have and p (x = x exp(x ( sinh x 4 n k= x k+3 k!(k + 3 ( n p (x = exp(x + 4x exp(x cosh x 4x ( ( = exp(x cosh x + 4x (exp(x It follows from the well-know inequalities: exp(x cosh ( x and exp(x = + x k /k! > k= n k= x k k! x k k! k=. n k= x k /k! that p (x. So p (x is increasing and since p ( =, for all x R, we have p(x p( =. This concludes the proof of Lemma 1. Proof of Proposition 3. Set g(x = exp(x f(x and q(x = exp(x f (x; n = g(x Q n (x. Let us recall that g (x and g (x have been determined in (3 and (4 respectively. We have n q (x = g (x 4x 5 x k ( 1 k!(k + 3(k + k x k + 7 and, by (4, k= q (x = g (x 4(1 + x x 4 n k= x k k!(k + 3(k + [ ( n = (1 + x exp(x cosh x x 4 k= ] x k. k!(k + 3(k + It follows from Lemma 1 that q (x, implying that q (x is increasing and since q ( =, for all x R, we have q(x q( =. So exp(x f (x; n. Since Q n (x for all x R and n N, it is clear that f (x; n f(x. The proof of Proposition 3 is complete. Proof of Proposition 4. Using x = ( x a + a x a and Proposition 1, we have exp(x = exp(( x a exp(a x a f( x a exp(a x a. Let us set v(x; a = f( x a exp(a x a. Hence we have exp(x max [f(x, v(x; a]. We will now show that v(x; a can be a better lower bound to f(x for some x. Let us study it according to the variable a. After several algebraic calculus and simplifications, we obtain a v(x; a = exp ( a x a w( x a, where w(y is the function defined by w(y = 1 [ ( exp y (3 + 18y + 1 ( y + 4y 3 + exp y ( y 1 ] y + 4y 3 1y, Since exp ( a x a >, we need to determine the sign of w( x a. Let us show that w(y is increasing on R. After several algebraic calculus and simplifications, we have ( w (y = exp y (6 + y 4y ( y 3 + exp y (6 y 4y + y 3 1, 1

12 ( w (y = exp y ( 5 1y ( y + 4y 3 + exp y (5 1y + y + 4y 3 and [ ( w (y = y exp y (3 + 8y ( y + exp y ( 3 + 8y + ] y. Let us observe that this last function is even and of the form w (y = y (R(y R( y, with R(y = exp ( y ( 3 + 8y + y. So it is enough to study its sign on [,, which corresponds to the sign of R(y R( y on [,. We have, for all y, [ ( (R(y R( y = exp y (1 6 ( y + y + exp y (1 + 6 ] y + y ( = 4(1 + y cosh y + 4 ( y sinh y. So R(y R( y is increasing on [,, we have R(y R( y R( R( = for all y, and, a fortiori, w (y = y (R(y R( y for all y. Since w (y is even, we have w (y for all y R. This implies that w (y is increasing on R. Since w ( =, w (y is decreasing (, ] and increasing on (,. So w (y w ( = implying that w(y is increasing on R. It follows from this result that, if a < x, we have w( x a > w( =, implying that av(x; a >, so v(x; a is strictly increasing according to a. If a > x, we have w( x a < w( =, implying that av(x; a <, so v(x; a is strictly decreasing according to a. Clearly, we have av(x; a = if and only if a = x. Since f(x is an even function and v(x; = v(x; x = f(x, for all x a/, we have exp(x = v(x; x = For x < a/, we have Hence, for all x R, we have sup v(x; a v(x; a inf v(x; a = inf [v(x;, v(x; x ] = f(x. a [, x ] a [, x ] f(x = inf v(x; a sup v(x; a v(x; a. a [, x ] a> x exp(x max [f(x, v(x; a] = f(x1 { x <a/} (x + v(x; a1 { x a/} (x = f (x; a f(x. Proposition 1 is proved. Proof of Proposition 5. Let us prove the two points in turn. Let us remark that, by two successive integrations, we have so (1 + t exp(t dt dy = exp(x = Owing to exp(x θ(x, we have Hence exp(x ω(x. exp(x ω(x = y exp(y dy = exp(x 1, (1 + t exp(t dt dy + 1. (5 (1 + t [ exp(t θ(t ] dt dy. 11

13 Let us define the function β(x by Then we have β (x = x β(x = x (1 + t θ(tdt dy + 1 θ(x. (1 + t θ(tdt θ (x, β (x = (1 + x θ(x θ (x. Thanks to the assumption (1 + x θ(x θ (x, we have β (x, implying that β (x is increasing. Since θ ( =, we have β ( = and the minimum of β(x is attained in x =. Since θ( = 1, for all x, we have β(x β( =, implying that β( x for all x R. Since θ(x is even, we have ω(x = (1 + t θ(tdt dy + 1 θ(x. The proof of Proposition 5 is complete. Proof of Proposition 6. We proceed as the proof of Proposition 5. If follows from the equality (5, exp(x θ 1 (x and exp(x θ (x that exp(x κ(x = Hence exp(x κ(x. Let us define the function φ(x by x φ(x = Two differentiations give φ (x = x [ (exp(t θ 1 (t + t (exp(t θ (t ] dt dy. (θ 1 (t + t θ (tdt dy + 1 θ 1 (x. (θ 1 (t + t θ (tdt θ 1(x, φ (x = (θ 1 (x + x θ (x θ 1 (x. Since (θ 1 (x + x θ (x θ 1 (x, we have φ (x, implying that φ (x is increasing. Since θ 1( =, we have φ ( = and the minimum of φ(x is attained in x =. Since θ( = 1, for all x, we have φ(x φ( =, implying that φ( x for all x R. Since θ(x is even, we have κ(x = By exchanging the role of θ 1 (x and θ (x, we obtain κ(x = This ends the proof of Proposition 6. Proof of Proposition 7. We have (θ 1 (t + t θ (tdt dy + 1 θ 1 (x. (θ 1 (t + t θ (tdt dy + 1 θ (x. exp(t [ (1 + t exp(t 1 ] dt = [ exp(t + t exp(t ] x = exp( x ( exp(x 1, 1

14 so exp(x = 1 + exp( x If exp(x θ(x for all x R, we have exp(x 1 + exp( x Proposition 7 is proved. exp(t [ (1 + t exp(t 1 ] dt. exp(t [(1 + tθ(t 1] dt = γ(x. References Bae, J. (7. Optimal polynomial lower bounds for the exponential function, Honam Math. J. 9, 4, Bae, J. and Kim, S. (9. On a generalization of an upper bound for the exponential function, J. of Math. Anal. and Appl. 353, 1, 1 7. Bagul, Y.J. (17. Inequalities involving circular, hyperbolic and exponential functions. Journal of Mathematical inequalities, 11, 3, Brenner, J. L. and Alzer, H. (1991. Proc. R. Soc. Edinburgh, Bullen, P. S. (1998. A Dictionary of Inequalities. Pitman Monographs and Surveys in Pure and Applied Mathematics, Vol. 97, Longman, Harlow. Chesneau, C. (17. On two simple and sharp lower bounds for exp(x, preprint. Kim, S. (7. Densely algebraic bounds for the exponential function, Proc. Amer. Math. Soc. 135, Kuang, J. (1993. Applied Inequalities, nd edition. Hunan Education Press, Chang- sha, China. Mitrinović, D. S. (1964. Elementary Inequalities. P. Noordhoff Ltd., Groningen. Qi, F. (1997. A method of constructing inequalities about e x, Univ. Beograd. Publ. Elektrotehn. Fak. Ser. Mat., 8, 16 3, (1997. Wolfram, S. (1999. The Mathematica Book (4th edition, Cambridge Univ. Pr., Cambridge. Also see, 13