Bounds for the generalized Marcum function of the second kind

In this paper, we focus on the generalized Marcum function of the second kind of order ν>0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu >0$$\end{document}, defined by Rν(a,b)=ca,νaν-1∫b∞tνe-t2+a22Kν-1(at)dt,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R_{\nu }(a,b)=\frac{c_{a,\nu }}{a^{\nu -1}} \int _b ^ {\infty } t^{\nu } e^{-\frac{t^2+a^2}{2}}K_{\nu -1}(at)\mathrm{d}t, \end{aligned}$$\end{document}where a>0,b≥0,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a>0, b\ge 0,$$\end{document}Kν\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{\nu }$$\end{document} stands for the modified Bessel function of the second kind, and ca,ν\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c_{a,\nu }$$\end{document} is a constant depending on a and ν\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu $$\end{document} such that Rν(a,0)=1.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{\nu }(a,0)=1.$$\end{document} Our aim is to find some new tight bounds for the generalized Marcum function of the second kind and compare them with the existing bounds. In order to deduce these bounds, we include the monotonicity properties of various functions containing modified Bessel functions of the second kind as our main tools. Moreover, we demonstrate that our bounds in some sense are the best possible ones.


Introduction and preliminary results
Let R ν (a, b) be the so-called generalized Marcum function of the second kind [4], defined by where a > 0, b ≥ 0, ν > 0, and K ν stands for the modified Bessel function of the second kind. As R ν (a, 0) = 1, we can write R ν (a, b) as follows: where a > 0, b ≥ 0, and ν > 0. In [4], Baricz et al. proved some monotonicity properties of the generalized Marcum function of the second kind with respect to a, b and ν, and log-concavity and convexity properties with respect to the parameter b. They also deduced several bounds for the Marcum function of the second kind (that is, when ν = 1) and for the generalized Marcum function of the second kind.
The study of the generalized Marcum function of the second kind is motivated by the importance of the generalized Marcum Q-function in the electrical engineering literature, which was studied by several authors in the last few decades, see for example [1,2,9] and the references therein. In particular, Baricz and Sun [2] obtained several tight bounds for the generalized Marcum Q-function which are based on the monotonicity properties of the functions of the form x → x αν+β I ν (x)/ f (e x , sinh x, cosh x), where α and β are real numbers and I ν stands for the modified Bessel function of the first kind. These bounds are sharp in nature in the sense that they cannot be further improved. Motivated by the work of Baricz and Sun [2], in this paper our aim is to find some new bounds for the generalized Marcum function of the second kind. These bounds are obtained from the monotonicity properties of the functions of the form x → x αν+β K ν (x)/ f (e x , sinh x, cosh x), where α and β are real numbers. The monotonicity properties of these functions depend upon some monotonicity properties of the ratio of modified Bessel functions of the second kind, which were investigated in [5] and [11]. The bounds which we deduce in this paper are tight enough and cannot be further improved as we prove that the power αν + β is the smallest or largest constant so that the corresponding monotonicity property for the function x → x αν+β K ν (x)/ f (e x , sinh x, cosh x) holds.
It is worth mentioning that during the preparation of our manuscript we found out that the distribution whose survival function we study in this paper resembles a distribution of Nadarajah [6]. More precisely, motivated by a Bayesian inference of an inverse Gaussian sample, Nadarajah [6] introduced a new modified Bessel distribution of the second kind supported on the real line. In [4], it has been shown that the generalized Marcum function of the second kind is in fact the survival function of the truncated distribution of a special case (when p = 1/2) of the modified Bessel distribution of the second kind considered by Nadarajah. Taking into account the vast application field in engineering and physical problems of the generalized Marcum Q-function and of the Bessel distributions, we believe that our paper may be useful for people who meet in their study such kind of exponential integrals involving the modified Bessel function of the second kind. This paper is organized as follows: in this section we prove the preliminary results which contain the monotonicity of the functions involving the modified Bessel function of the second kind. Using the preliminary results in Sect. 2, we present some new lower and upper bounds for the generalized Marcum function of the second kind for two cases: when b ≥ a > 0 and a > b > 0. In Sect. 3, we provide some comparison between our new bounds which are obtained in Sect. 2 and the bounds given in [4] for both cases when b ≥ a > 0 and a > b > 0. The discussion is concluded in Sect. 4, by illustrating the theoretical results numerically.
It is interesting to note that the functions x → x ν K ν (x)e x and x → x −ν K ν (x)e x have different monotonic behavior, see [4]. More specifically Moreover, Yang and Zheng [11] showed that for fixed ν ≥ 1/2 the function x → √ x K ν (x)e x is decreasing on (0, ∞). Then the natural question which we can ask is that for a fixed ν what is the monotonic behavior of the function x → x α ν K ν (x)e x on (0, ∞)? Does there exist the smallest α ν such that the function x → x α ν K ν (x)e x is increasing on (0, ∞)? Does there exist the largest α ν such that the function x → x α ν K ν (x)e x is decreasing on (0, ∞)? These questions will be answered in this section. Moreover, this study improves one of the results of Yang and Zheng [11] and provides an alternative proof for [10,Corollary 1]. Similarly, the behavior of the functions x → x β ν K ν (x)/ sinh x and x → x β ν K ν (x)/ cosh x are studied and we find the largest and the smallest powers of x such that these functions possess the monotonic decreasing/increasing property. This section also discusses the monotonic behavior of the functions The following two lemmas are used to prove the main Lemma in this section. For Lemma 1 we refer [11,Theorem 2.6], while for Lemma 2 we refer to [3,Theorem 2] and [11,Lemma 2.4].
The main Lemma in this section is the following result.

Lemma 3
Let ν ∈ R. Then the following assertions are true: Then In view of [7, 10.29 To find the smallest value of α ν for which the function f ν is increasing on (0, ∞), it is necessary to find the minimum of the function φ ν . By Soni inequality [8,Eq. 5] and the fact K ν (x) = K −ν (x), we have Now, in view of the asymptotic formula [7, 10.30.2] where ν > 0 and x → 0, we get lim x→0 φ ν (x) = 0. (1.8) By using (1.6) and (1.8) we obtain f ν (x) ≥ 0 for α ν ≥ ν. Thus, α ν = ν is the smallest α ν for which x → x α ν K ν (x)e x is increasing on (0, ∞). Now, our aim is to find the largest constant α ν for which the function x → x α ν K ν (x)e x is decreasing on (0, ∞) for all ν ≥ 1/2. We again rewrite the derivative of f ν (x) as follows: . Thus, to find the largest constant for which the function f ν is decreasing it is necessary to find the maximum value of the function ψ ν . Due to Lemma 1, we have that for any x > 0 and ν ≥ 1/2. Now, by the recurrence relation (1.3) and using the fact that Thus for all x > 0 we get By using once again the recurrence relation (1.3) and in view of the asymptotic expansion [7, 10.40.2] (1.12) By using (1.11) and (1.12), it follows that f (x) ≤ 0 for all x ∈ (0, ∞) when α ν ≤ 1/2. Thus α ν = 1/2 is the largest value for which the function From Lemma 2 and the monotone decreasing property of Thus from (1.13) we obtain and consequently for all Observe that For all ν ∈ R proceeding like in part c, the function Thus (1.14) becomes Hence for all γ ν ≤ ν the function e. First note that (1.16) From (1.5) and coth x > 1, we have Consequently for ν > 1/2 and x > 0 we have (1.18) and the inequality (1.17) follows from (1.5). It is easy to check that the function φ has maximum value at On the other hand, by using the inequality (1.10), we have In view of one can conclude that for ν ≥ 1/2 the largest constant for which the function x →

Remark 1
Recall that the functions given in part c of [4, Lemma 5] and part a of Lemma 3 are the sharpest in terms of their monotonicity. From part b we can say that these functions cannot be further improved by raising or diminishing the power of x in the functions. As a consequence of part b of Lemma 3, we can conclude that if 1/2 < α < ν, then the function has at least one real positive zero. For example, using the expression

Remark 2
It is worth mentioning that recently Yang and Tian [10, Corollary 1] obtained the result of part b of Lemma 3 by studying the properties of ratios of two Laplace transforms. In this paper an independent proof is provided.
x coth x → −∞ whenever x → ∞, by using (1.13), we conclude that the function x → x β ν K ν (x)/ sinh x is not an increasing function on the whole interval (0, ∞) for any β ν . Similarly, x → x γ ν K ν (x)/ cosh x is not an increasing function on the whole interval (0, ∞) for any γ ν . Recall also that for a, ν > 0, c a,ν denotes the constant defined by Theorem 1 For b ≥ a > 0, the following inequalities hold: Proof For ν ≥ 1/2 and t ≥ b part a of Lemma 3 leads to which in view of (1.1) implies that To prove (2.2), the monotone property of x → x ν+1 K ν (x)/ sinh x is used. Note that part c of Lemma 3 implies for all t ≥ b and ν ∈ R, and this implies that Similarly, from part d of Lemma (3) we get for all t ≥ b and ν ∈ R. By using the above inequality in view of (1.1) we get Theorem 2 For ν > 3/2 and b ≥ a > 0, the following inequalities hold true: Using part e of Lemma 3 with constant δ ν = ν and proceeding like in the previous theorem we can deduce inequality (2.7). Similarly, part f of Lemma 3 with ν = ν + 1 yields the inequality (2.8) since part f of Lemma 3 is true for all ν ≥ ν + 0.278 . . .. The choice ν = ν + 1 is just sake of convenience.

Remark 4
Using the well-known inequality 1 1−e −2at > 1 for all a, t > 0, we can obtain from the lower bound (2.7) a weaker lower bound, which can be expressed in terms of the complementary error function as follows: The best choice for ν in part f of Lemma 3 is ν = ν + θ where θ = 0.278 . . . which gives a tighter lower bound for R ν (a, b) than the lower bound given in (2.8), that is precisely
Proof For ν > 1/2 and 0 < t ≤ b, part a of Lemma 3 implies The inequality (2.11) is obtained by using the monotonicity of the function t → t ν+1 K ν (t)/sinh t. For 0 < t ≤ b and ν ∈ R, part c of Lemma 3 leads to (2.14) Using the above inequality in view of (1.2), for ν > 0 we get Similarly, by using part d of Lemma (3) we obtain the upper bound for R ν (a, b) given in (2.12).
Theorem 4 For ν > 3/2 and a > b > 0, the following inequalities hold:   R ν (a, b) a weaker lower bound, which can be expressed in terms of complementary error function as follows:

Sharpness of the bounds and comparison with other existing bounds
In this section, we discuss the tightness of the bounds obtained in Sect. 2 and compare these bounds with the bounds given in [4]. All the bounds for R ν (a, b) stated in the previous section and the section 2 of [4] are obtained by using the bounds for K ν (t).
Thus we compare the bounds for K ν (t), which give immediately the comparison of the bounds for R ν (a, b).

Comparison with other existing bounds
Let b ≥ a > 0. For ν ≥ 1/2, Baricz et al. [4] obtained the following upper bound for by using the inequality In this paper, by using the inequality for ν ≥ 1/2, a new upper bound is obtained for R ν (a, b) in the case when ν ≥ 3/2 On the other hand, for ν > 3/2 we have Hence the new bound (3.2) is sharper than the upper bound (3.1) obtained by Baricz et al. [4] for ν > 3/2. For ν > 0, Baricz et al. [4] obtained the following upper bound for R ν (a, b) by using the monotone decreasing property of the function t → t −ν e −t K ν (t) on (0, ∞) and for t ≥ b the inequality Note that t → e −t cosh t and t → t coth t are decreasing and increasing functions, respectively, on (0, ∞).
Consequently for t ≥ b and ν > 0 In view of (3.5), (2.5), and (2.6), we conclude that the upper bound in (2.2) is sharper than the upper bound in (2.3). Moreover, in view of (3.4) and (3.5), we conclude that the new bound (2.3) is sharper than the upper bound in (3.3) obtained by Baricz et al. [4].

Remark 6
In view of (3.5) for ν > 0 and t ≥ b > 0, we get By using this we get the following new upper bound for R ν (a, b) In view of (3.5), we conclude that the new upper bound (3.6) is sharper than the upper bound (3.3) obtained by Baricz et al. [4].
For ν ≥ 3/2, Baricz et al. [4] obtained the following lower bound for R ν (a, b) by using the monotone increasing property of the function t → e t t ν K ν (t) on (0, ∞) for all ν ≥ 1/2, that is, for t ≥ b > 0 and ν ≥ 1/2 the inequality Note that t → e −t sinh t and t → t coth t both are increasing functions on (0, ∞). Consequently, In view of (3.9), the new lower bound in (2.7) is sharper than the new lower bound in (2.8). Moreover, in view of (3.9) and (3.8), we conclude that the lower bound in (3.7) obtained by Baricz et al. [4] is sharper than the new lower bound in(2.7). It is easy to see that the lower bound in (2.9) is sharper than the lower bound in (2.8) but weaker than the lower bound in (3.7) which is obtained by Baricz et al. [4].
Similarly to the previous discussion, for the case when a > b > 0 we can verify the followings: ii. The upper bound in (2.10) is sharper than the upper bound in (3.11) given in [4] iii. The lower bound in (2.15) is sharper than the lower bound in (2.16), but weaker than the lower bound in (3.12) obtained by Baricz et iv. The lower bound in (2.17) is sharper than the lower bound in (2.16) but weaker than the lower bound in (3.12).

Remark 7
By using the monotone decreasing property of the functions t → t −ν e −t K ν (t), t → e −t cosh t and the monotone increasing property of the function t → t cosh t we get By comparing the first and the last one, we get the following new upper bound for ν > 0, Clearly, the new upper bound in (3.14) is sharper than the upper bound in (3.10) obtained by Baricz et al. [4].

Tightness of the bounds for b → 0 and b → ∞
In this subsection, the tightness of the new upper and lower bounds are discussed. Section 3.1 shows that all the new upper bounds obtained in this study are sharper than the upper bounds obtained by Baricz et al. [4]. Moreover, the upper bounds obtained by Baricz et al. [4] are tight bounds. Consequently, all the new upper bounds obtained in this study are also tight. It is also possible to verify that the relative error for the upper bounds tend to zero as b approaches infinity. In this subsection, proofs are discussed to show that all the new lower bounds are tight as well as the relative error in the lower bounds tends to zero as b → ∞.
The following limits show the tightness of the relative error for the lower bounds: (3.16) Using the limit (3.15), we conclude that the lower bound L 2 in (2.8) tends to zero as b → ∞. Moreover, by using (3.16) and the following asymptotic relation we can verify that the relative error (L 2 − R ν (a, b))/R ν (a, b) tends to zero as b → ∞.
As the lower bound L 1 in (2.7) is sharper than the lower bound L 2 in (2.8), L 1 is also tight as b → ∞. Similarly, we can verify that the relative error in L 1 approaches to zero as b → ∞. Similarly, for the case a > b > 0 all the new lower bounds (2.15) and (2.16) are also tight as well as the relative error tends to zero as b → 0.

Numerical results
In this section, we present numerical results in order to compare the various upper and lower bounds of R ν (a, b). To draw the figures for the bounds of R ν (a, b) we have used Mathematica 8.0. For the choice ν = 3 and a = 1, Fig. 1 illustrates the comparison between all the upper bounds with R 3 (1, b) in the interval (1, 2.5). Let R 3 denote R 3 (1, b). Figure 1 supports the theoretical results obtained in Sect. 3. Note that the new upper bound U 1 is sharper than all the other upper bounds. For large b, the upper bound U 5 is sharper than the upper bound U 3 .
For the choice ν = 3 and a = 1, Fig. 4 illustrates the comparison between all the lower bounds with R 3 (1, b) in the interval (0, 1). Let R 3 denote R 3 (1, b). Figure 4 supports the theoretical results obtained in Sect. 3.
Funding Open access funding provided by Óbuda University.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.