Monotonicity properties of q -digamma and q -trigamma functions

Monotonicity properties of q -digamma and q -trigamma functions

Available online at www.sciencedirect.com ScienceDirect Journal of Approximation Theory 192 (2015) 336–346 www.elsevier.com/locate/jat Full length a...

197KB Sizes 1 Downloads 144 Views

Available online at www.sciencedirect.com

ScienceDirect Journal of Approximation Theory 192 (2015) 336–346 www.elsevier.com/locate/jat

Full length article

Monotonicity properties of q-digamma and q-trigamma functions Necdet Batir Department of Mathematics, Faculty of Sciences and Arts, Nevs¸ehir Haci Bektas¸ Veli University, Nevs¸ehir, Turkey Received 21 May 2014; received in revised form 23 December 2014; accepted 29 December 2014 Available online 14 January 2015 Communicated by Paul Nevai

Abstract Some complete monotonicity results for q-polygamma functions are proved. Our results extend positivity of some functions containing q-polygamma functions to complete monotonicity property. Also, we give two new inequalities for q-trigamma function. c 2015 Elsevier Inc. All rights reserved. ⃝

MSC: primary 33B15; 26D15 Keywords: q-digamma function; q-psi function; q-trigamma function; q-gamma function; q-extensions; Inequalities

1. Introduction As it is known, the gamma function 0(x) is defined by the improper integral  ∞ 0(x) = t x−1 e−t dt, x > 0. 0

The most important function related to the gamma function is the digamma or psi function ψ, ′ (x) . The derivatives which is defined as the logarithmic derivative of Γ , namely ψ(x) = Γ0(x) ′ ′′ ψ (x), ψ (x), . . . are known to be the polygamma functions in the literature. Particularly ψ ′ and E-mail addresses: [email protected], [email protected]. http://dx.doi.org/10.1016/j.jat.2014.12.013 c 2015 Elsevier Inc. All rights reserved. 0021-9045/⃝

337

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

ψ ′′ are called the trigamma and tetragamma functions, respectively. The q-analogue of 0(x), denoted by Γq (x), was introduced by Jackson [16] as Γq (x) = (1 − q)1−x

∞  1 − q n+1 , 1 − q n+x n=0

0
(1.1)

and Γq (x) = (q − 1)1−x q x(x−1)/2

∞  1 − q −(n+1) , 1 − q −(n+x) n=0

q>1

(1.2)

for x > 0. It was proved in [19] that lim Γq (x) = lim Γq (x) = 0(x).

q→1−

q→1+

Γ ′ (x)

Similarly the q-digamma or q-psi function ψq (x) is defined by ψq (x) = Γqq (x) . The derivatives ψq′ , ψq′′ , . . . are called the q-polygamma functions. In particular the functions ψq′ and ψq′′ are called q-trigamma and q-tetragamma functions, respectively. In [17] it was shown that limq→1− ψq (x) = limq→1+ ψq (x) = ψ(x). From the definitions (1.1) and (1.2) one can easily deduce that ψq (x) = − log(1 − q) + log q

∞  q nx , 1 − qn n=1

0
(1.3)

and 

∞ 1  q −nx ψq (x) = − log(q − 1) + log q x − − 2 n=1 1 − q −n

 ,

q > 1.

Differentiation of (1.3) and (1.4) gives  ∞  nq nx  2   log q ; 0 < q < 1,   1 − qn n=1 ′ ψq (x) = ∞   nq −nx  2  log q + log q ; q > 1.   1 − q −n n=1

(1.4)

(1.5)

Alzer and Grinshpan [3] proved that ψq′ is completely monotonic on (0, ∞), that is, (−1)n (ψq′ (x))(n) > 0,

x > 0, q > 0, n = 0, 1, 2, . . . .

(1.6)

We recall that a function f is said to be completely monotonic on an interval I if f has derivatives of all orders on I and (−1)n f (n) (x) ≥ 0 for all x ∈ I and all integers n ≥ 0. These functions have important applications in probability and numerical analysis. In particular, completely monotonic functions involving the gamma and q-gamma functions are very important because they enable us to estimate the polygamma and q-polygamma functions. For more information for these functions, we refer to Chapter IV of [29]. A short calculation gives for x > 0 and q > 0 Γq (x + 1) =

1 − qx Γq (x) 1−q

(1.7)

338

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

and Γq (x) = q

(x−1)(x−2) 2

Γq −1 (x).

(1.8)

If we take logarithm of both sides of (1.7) and (1.8) and then differentiate, we find ψq (x + 1) − ψq (x) =

q x · log q , qx − 1

ψq′ (x + 1) − ψq′ (x) = − ψq′′ (x + 1) − ψq′′ (x) =

(1.9)

q −x · log2 q , (1 − q −x )2

q −x · (1 + q −x ) log3 q , (1 − q −x )3

(1.10) (1.11)

  3 ψq (x) = x − log q + ψ1/q (x), 2

(1.12)

′ ψq′ (x) = log q + ψ1/q (x),

(1.13)

and

′′ ψq′′ (x) = ψ1/q (x).

We refer [3,1,4,12,14,15,18] for basic properties of the q-gamma and q-digamma functions. Recently, many monotonicity and complete monotonicity properties and inequalities for the gamma and digamma functions have been extended to the q-gamma and q-digamma functions; see, for example, [3,1,9–11,20,23,24,26,27,25,22] and references therein. In [2] H. Alzer proved  2 ψ ′′ (x) + ψ ′ (x) > 0

(1.14)

for x > 0. The author rediscovered it in [5] and used it to prove interesting inequalities for the digamma function, see [7,6,5]. Alzer and Grinshpan [3] obtained a q-analogue of (1.14) and proved that  2 ψq′′ (x) + ψq′ (x) > 0

(1.15)

for q > 1 and x > 0. In [20], F. Qi showed that the function given in (1.15) is completely monotonic for q > 1 on (0, ∞). The author [8, Lemma 1] provided another q-extension of (1.14) and proved that 2  ψq′′ (x) + ψq′ (x) − (log q) · ψq′ (x) > 0

(1.16)

for all q > 0 and x > 0. Note that putting q = 1 in (1.16) one obtains (1.14). The inequality given in (1.16) played a central role in the proofs of inequalities in [8]. Our first aim in this work is to show the function in (1.16) is completely monotonic for all q > 0 on (0, ∞). The author [8, Theorem 2.1] proved that the function Fq defined by 1

1 − q x+ 2 Fq (x) = ψq (x + 1) − log 1−q

is positive for q > 1. Our second aim is to prove that Fq (x) is completely monotonic for q > 0 on (0, ∞).

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

339

In [5] the author proved that 1

1 − e− x +

1 1 < ψ ′ (x) < e x − 1. 2 x

(1.17)

In [13] the authors employed the second inequality in (1.17) to improve some inequalities for polygamma functions. In [21] the authors proved that the function f (x) = e1/x − ψ ′ (x) is completely monotonic on (0, ∞) and applied this to bound the modified Bessel function I p (x). Our final aim in this work is to obtain a q-analogue of (1.17). 2. Main results The following theorems are our main results. Theorem 2.1. Let q > 0 and  2 G q (x) = ψq′′ (x) + ψq′ (x) − log q · ψq′ (x). Then G q is completely monotonic on (0, ∞). Proof. We shall use an idea from [3]. Let q > 1. Then by [8] G q (x) > 0 for q > 0 and x > 0. So, to prove Theorem 2.1, we only need to show (−1)k G q(k) (x) > 0

(2.1)

for k = 1, 2, 3, . . . , and x, q > 0. Applying (1.5) we obtain G q (x) log2 q

=

∞  i−1 

c j (q, x)ci− j (q, x) −

i=2 j=1

=

 ∞  i−1  i=2

∞ 

(i − 1)ci (q, x)

i=2

 c j (q, x)ci− j (q, x) − (i − 1)ci (q, x) ,

j=1

where ci (q, x) =

i · q −i x · log q . 1 − q −i

Hence, in order to prove (2.1) it is enough to see (−1)k

d k G q (x) > 0, d x k log2 q

or equivalently (−1)k

∞  i−1 k  ∂ [c j (q, x)ci− j (q, x)] i=2 j=1

∂xk

≥ (−1)k

∞  ∂ k [ci (q, x)] (i − 1) . ∂xk i=2

If we take the partial derivatives here, we see that this is implied by i−1  j=1

(q j

j (i − j) i(i − 1) ≤ , i− j − 1)(1 − q ) (log q)(1 − q i )

i ≥ 2,

(2.2)

340

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

which is proved in [3]. This proves that G q is completely monotonic for q > 1 on (0, ∞). But if we use (1.13) we see that G q (x) = G 1/q (x), and therefore G q is completely monotonic for q > 0 on (0, ∞).  I would like to add another proof of Theorem 2.1 which was provided by an anonymous referee. Second proof of Theorem 2.1. Let q > 1. Applying (1.10) and (1.11) we get G q (x + 1) − G q (x) =

q −x log2 q · g(x), (1 − q −x )2

where g(x) = −2ψq′ (x) + log q +

q −x · log2 q + (1 − q −2x ) · log q . (1 − q −x )2

By using (1.10), easy calculations give g(x + 1) − g(x) =

log q (1 − q −x )2 (1 − q −x−1 )2

· h(x)

(2.3)

where q 2 · h(x) = (2q − 2q 2 + q(q + 1) log q) · q −x + (2q 2 − 2 − 4q · log q) · q −2x + (2 − 2q + (q + 1) · log q) · q −3x . h is completely monotonic on (0, ∞) as linear combination with positive coefficients of expressions exp(−kx log q) with k = 1, 2, 3. Then (2.3) shows that g(x + 1) − g(x) is completely monotonic because 1 − q −x = 1 − exp(−x log q) is a Bernstein function and then the reciprocal (and its positive powers) are completely monotonic, see [28] or [29]. Also, g(x + n + 1) − g(x + n) is completely monotonic for n = 0, 1, 2, . . . and the identity g(x + n) − g(x) =

n−1  [g(x + k + 1) − g(x + k)] k=0

gives that g(x + n) − g(x) is completely monotonic. Finally for n → ∞ we conclude that −g(x) is completely monotonic, i.e., G q (x) − G q (x + 1) is completely monotonic on (0, ∞) for q > 1. The same procedure as for g proves that G q (x) is completely monotonic. From the identity G q (x) = G 1/q (x) it follows that G q (x) is completely monotonic on (0.∞) for all q > 0, which was the assertion of Theorem 2.1. We recall that a function f : I ⊆ (−∞, ∞) → [0, ∞) is called a Bernstein function on I if f (t) has derivatives of all orders and f ′ (t) is completely monotonic on I ; see [28]. Theorem 2.2. Let q > 0 and Fq be defined by   1 1 − q x+ 2 . Fq (x) = ψq (x + 1) − log 1−q Then Fq is completely monotonic on (0, ∞).

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

341

Proof. In Theorem 2.1 of [8] it was proved that Fq (x) > 0 for q > 0 and x > 0. Now let q > 1. Differentiation of Fq (x) gives Fq′ (x) = ψq′ (x + 1) −

log q 1

1 − q −x− 2

.

Since ψq′ (x + 1) = log q + log2 q

∞  n · q −n(x+1)

1 − q −n

n=1

,

and 1 1

1 − q −x− 2

=1+

∞ 

q −n(x+1/2)

n=1

we get Fq′ (x) = − log q

∞  q −3n/2 an (q) n=1

1 − q −n

· e−nx log q ,

(2.4)

where an (q) = q n − 1 − n · q n/2 · log q. It is very easy to show that an (q) is positive for all n ∈ N. From Eq. (2.4) together with this fact we conclude that −Fq′ (x) is completely monotonic, which necessarily implies that Fq (x) is completely monotonic for q > 1 since Fq (x) is positive for x > 0 [8]. Employing the first identity in (1.13) we arrive at Fq (x) = F1/q (x) for all x > 0 and q > 0, that is, Fq is completely monotonic on (0, ∞) for q > 0.  Corollary 2.3. Let x and q be positive real numbers. Then the following inequalities hold: 1

1

1 − q x+ 2 1 − q x+ 2 α∗ + log < ψq (x + 1) < α ∗ + log , 1−q 1−q √ where α∗ = 0 and α ∗ = ψq (1) + log(1 + q) are the best possible constants, and β∗ +

log q 1−q

−x− 12

< ψq′ (x + 1) < β ∗ +

log q 1

1 − q −x− 2

,

(2.5)

(2.6)

where q log q β∗ = ψq′ (1) + √ q −q

and

β∗ = 0

are the best possible constants. Proof. By Theorem 2.2 Fq is strictly decreasing and Fq′ is strictly increasing. From (1.3) and (1.5) we find lim ψq (x) = − log(1 − q),

x→∞

0 < q < 1,

(2.7)

and lim ψq′ (x) = log q,

x→∞

q > 1.

(2.8)

342

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

Let 0 < q < 1. Then, by using (2.7) we obtain   lim Fq (x) = lim

x→∞

x→∞

1

1 − q x+ 2 ψq (x + 1) − log 1−q



= − log(1 − q) + log(1 − q) = 0. If q > 1, then by using (1.12) and (2.7), we obtain      1 1 1 − q x+ 2 log q − log lim Fq (x) = lim ψ1/q (x + 1) + x − x→∞ x→∞ 2 1−q   1−q = − log(1 − 1/q) + lim log = 0. x→∞ q −x+1/2 − q

(2.9)

(2.10)

Combining monotonic decrease of Fq with (2.9), (2.10) and (2.8) we obtain 0 = α∗ = lim Fq (x) < Fq (x) < Fq (0) = α ∗ , x→∞

which is equivalent to (2.5). Now let us prove (2.6). If q > 1 applying (2.8) we get   log q ′ ′ = 0. lim F (x) = lim ψq (x + 1) + 1 x→∞ q x→∞ q −x− 2 − 1 If 0 < q < 1, then 1/q > 1, and hence applying (1.13) and (2.8) yields   log q ′ ′ lim F (x) = lim log q + ψ1/q (x + 1) + = 0. 1 x→∞ q x→∞ q −x− 2 − 1

(2.11)

(2.12)

Taking into account monotonic increase of Fq′ , and the limit values in (2.11) and (2.12), we get for q > 0 and x ≥ 0 β∗ = Fq′ (0) < Fq′ (x) < lim Fq′ (x) = β ∗ = 0, x→∞

which is equivalent to (2.6).



Theorem 2.4. Let q > 0 and x > 0. Then log q log q < ψq′ (x + 1) < , α−x 1−q 1 − q β−x where the constants   log q 1 · log 1 − ′ α= log q ψq (1) and    1 q log q   − · log ; if 0 < q < 1,  log q q −1   β= 1 log q    · log ; if q > 1 log q q −1 are the best possible.

(2.13)

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

343

Proof. Let us define   1 log q Hq (x) = x + · log 1 − ′ . log q ψq (x + 1)

(2.14)

Hq is a well-defined function because the expression under log is positive. This is clear for 0 < q < 1, and for q > 1 because ψq′ (x) > log q by (2.8) and the fact that ψq′ is decreasing. First we shall show that Hq is strictly increasing on (0, ∞) for q > 0. Differentiation of Hq yields  2 ψq′′ (x + 1) + ψq′ (x + 1) − log q · ψq′ (x + 1) . Hq′ (x) =  2 ψq′ (x + 1) − log q · ψq′ (x + 1) Utilizing (1.13) we see that  2 ′ (x + 1) > 0. ψq′ (x + 1) − log q · ψq′ (x + 1) = ψq′ (x + 1)ψ1/q Hence, by Theorem 2.1 we get Hq′ (x) > 0 for x > 0 and q > 0. Secondly, we need to show the limit values    1 q log q   · log ; if 0 < q < 1, − log q q −1   lim Hq (x) = (2.15) x→∞ 1 log q    · log ; if q > 1. log q q −1 Let 0 < q < 1. Then we get 1 · log lim Hq (x) = x→∞ log q



 lim

x→∞

log q q − −x−1 q ·q ψq′ (x + 1) x

From (1.5) we can easily deduce that  2   log q ; if 0 < q < 1, lim q −x ψq′ (x) = 1 − q x→∞  0; if q > 1, consequently we get for 0 < q < 1   q log q 1 · log . lim Hq (x) = − x→∞ log q q −1 Now let q > 1. Then upon using (1.13) we see that   ′ (x + 1) (1/q)−x−1 ψ1/q 1 lim Hq (x) = . · log lim x→∞ x→∞ q log q + qψ ′ (x + 1) log q 1/q ′ (x + 1) = 0, and Since limx→∞ ψ1/q ′ lim (1/q)−x−1 ψ1/q (x + 1) =

x→∞

log2 (1/q) 1 − 1/q

 .

(2.16)

344

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

by (2.16), we get   1 log q . lim Hq (x) = · log x→∞ log q q −1 This proves (2.15). Combining monotonic increase of Hq with (2.15) we get for all q > 0 and x >0   1 log q α= · log 1 − ′ = Hq (0) < Hq (x) < lim Hq (x) = β (2.17) x→∞ log q ψq (1) or 1 1 − q α−x 1 − q β−x < ′ < , log q ψq (x + 1) log q which is equivalent to (2.13).



Our last theorem provides a q-analogue of (1.17). Theorem 2.5. Let x > 0. Then the following inequalities hold: If q > 1     log q q −x · log2 q log q ′ 1 − exp − + < ψq (x) < exp −1 1 − q −x 1 − q −x (1 − q −x )2

(2.18)

and if 0 < q < 1  q x · log2 q log q + 1 + log q − exp < ψq′ (x) 1 − qx (1 − q x )2   log q < −1 + log q + exp − . 1 − qx 

(2.19)

Proof. Let q > 1. Applying the mean value theorem to eψq (t) on the interval [x, x + 1], we get eψq (x+1) − eψq (x) = ψq′ (x + δ)eψq (x+δ) ,

0 < δ < 1.

(2.20)

We define u(x) = ψq′ (x)eψq (x) , x > 0, q > 1. Differentiation gives u ′ (x) = {ψq′′ (x) + [ψq′ (x)]2 }eψq (x) . By (1.15) u is strictly increasing. Since     log q exp{ψq (x + 1)} − exp{ψq (x)} = exp{ψq (x)} exp −1 1 − q −x by (1.9), we conclude from (2.20) that     log q ′ ψq (x) ψq (x) ψq (x)e
N. Batir / Journal of Approximation Theory 192 (2015) 336–346

345

Remark 2.7. Letting q → 1 in Theorem 2.2 leads to the observation that F1 (x) = ψ(x + 1) − log(x + 1/2) is completely monotonic on (0, ∞). For 0 ≤ α one finds easily the integral representation  ∞ f α (x) = ψ(x + 1) − log(x + α) = e−xt δα (t)dt, 0

with δα (t) =

exp((1 − α)t) − exp(−αt) − t t (exp(t) − 1)

and one observes that δα (t) > 0 for t > 0 if and only if 0 ≤ α ≤ 1/2. In particular one has the integral representation of the completely monotonic function F1 . Remark 2.8. If we let q → 1 in (2.17), we obtain −

1 ψ ′ (1)

= H1 (0) < H1 (x) = x −

1 1 < lim H1 (x) = − x→∞ + 1) 2

ψ ′ (x

or 1 1 < ψ ′ (x + 1) < , x + 1/2 x + 6/π 2 since ψ ′ (1) = π 2 /6, which is a new inequality as far as we know. Acknowledgments I would like to gratefully and sincerely thank the anonymous referees for their very valuable comments and useful corrections on this work, which improved the quality of the paper significantly. Furthermore, I thank very much an anonymous referee for providing the second proof of Theorem 2.1 and the remarks given in Remark 2.7. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

H. Alzer, Sharp bounds for q-gamma functions, Math. Nachr. 222 (2001) 5–14. H. Alzer, Sharp inequalities for the digamma and polygamma functions, Forum Math. 16 (2) (2004) 181–221. H. Alzer, A.Z. Grinshpan, Inequalities for the gamma and q-gamma functions, J. Approx. Theory 144 (2007) 67–83. R. Askey, The q-gamma and q-beta functions, Appl. Anal. 8 (1978) 125–141. N. Batir, Some new inequalities for gamma and polygamma functions, J. Inequal. Pure Appl. Math. 6 (4) (2005) Art. 103. Available online at: http://www.emis.de/journals/JIPAM/article577.html. N. Batir, On some properties of digamma and polygamma functions, J. Math. Anal. Appl. 328 (2007) 452–465. N. Batir, Inequalities for the gamma function, Arch. Math. (Basel) 91 (2008) 554–563. N. Batir, q-Extensions of some estimstes associated with the digamma function, J. Approx. Theory 174 (2013) 54–64. C. Berg, H.B. Petersen, On the iteration leading to a q-analogue of the digamma function, Fourier Anal. Appl. 19 (4) (2013) 762–776. P. Gao, Some monotonicity properties of gamma and q-gamma functions, ISRN Math. Anal. (2011) Article ID 375715, 15 pages. P. Gao, Some completely monotonic functions involving the q-gamma function, Math. Inequal. Appl. 17 (2) (2014) 451–460. A.Z. Grinshpan, M.E.H. Ismail, Completely monotonic functions involving the gamma and q-gamma functions, Proc. Amer. Math. Soc. 134 (2006) 1153–1160. B.-N. Guo, F. Qi, Improvement of lower bound of polygamma functions, Proc. Amer. Math. Soc. 141 (2013) 1007–1015.

346

N. Batir / Journal of Approximation Theory 192 (2015) 336–346

[14] M.E.H. Ismail, L. Lorch, M.E. Muldoon, Completely monotonic functions associated with the gamma and its q-analogues, J. Math. Anal. Appl. 116 (1986) 1–9. [15] M.E.H. Ismail, M.E. Muldoon, Inequalities and monotonicity properties for the gamma and q-gamma functions, in: R.V.M. Zahar (Ed.), Approximation and Computation, in: International Series of Numerical Mathematics, vol. 119, Birkh¨auser, Boston, M.A, 1994, pp. 309–323. [16] F.H. Jackson, On q-definite integrals, Quart. J. Pure Appl. Math. 41 (1910) 193–203. [17] C. Krattenthaler, H.M. Sirivastava, Summations for basic hypergeometric series involving a q-anologue of the digamma function, Comput. Math. Appl. 32 (2) (1996) 73–91. [18] T. Mansour, A.SH. Shabani, Some inequalities for q-digamma function, J. Inequal. Pure Appl. Math. (JIPAM) 10 (1) (2009) Article 12. [19] D.S. Moak, The q-gamma function for q > 1, Aequationes Math. 20 (1980) 278–288. [20] F. Qi, Some completely monotonic functions involving the q-tri-gamma and q-tetra-gamma functions with applications, Rev. R. Acad. Cienc. Exactas F´ıs. Nat. Ser. A Mat. (2015). http://dx.doi.org/10.1007/s13398-0140193-3. in press. [21] F. Qi, C. Berg, Complete monotonicity of a difference between the exponential and trigamma functions and properties related to a modified Bessel function, Mediterr. J. Math. 10 (4) (2013) 1685–1696. [22] A. Salem, Completely monotonic functions related to the q-gamma and the q-trigamma functions, Anal. Appl. (2015). http://dx.doi.org/10.1142/S0219530514500195. in press. [23] A. Salem, A completely monotonic function involving q-gamma and q-digamma functions, J. Approx. Theory 164 (2012) 971–980. [24] A. Salem, An infinite class of completely monotonic functions involving the q-gamma function, J. Math. Anal. Appl. 406 (2) (2013) 392–399. [25] A. Salem, Some properties and expansions associated with the q-digamma function, Quaest. Math. 36 (2013) 67–77. [26] A. Salem, Complete monotonicity properties of functions involving q-gamma and q-digamma functions, Math. Inequal. Appl. 17 (3) (2014) 801–811. [27] A. Salem, Two classes of bounds for q-gamma and q-digamma functions in terms of q-zeta functions, Banach J. Math. Anal. 8 (1) (2014) 109–117. [28] R. Schilling, R. Song, Z. Vondraˇcek, Bernstein Functions: Theory and Applications, de Gruyter, Berlin, 2010. [29] D.V. Widder, The Laplace Transform, Princeton University Press, 1946.