Abstract
In the article, we establish an inequality for Csiszár divergence associated with s-convex functions, present several inequalities for Kullback–Leibler, Renyi, Hellinger, Chi-square, Jeffery’s, and variational distance divergences by using particular s-convex functions in the Csiszár divergence. We also provide new bounds for Bhattacharyya divergence.
Similar content being viewed by others
1 Introduction
A real-valued function \(\psi : I \rightarrow \mathbb{R}\) is said to be convex if the inequality
holds for all \(\xi , \zeta \in I\) and \(\alpha , \beta \geq 0\) with \(\alpha +\beta =1\). It is well known that \(\psi : I \rightarrow \mathbb{R}\) is convex if and only if
for all \(\xi _{i}\in I\) and \(\alpha _{i}\geq 0\) with \(\sum_{i=1}^{n} \alpha _{i}=1\).
Convex function has wide applications in pure and applied mathematics, physics, and other natural sciences [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20]; it has many important and interesting properties [21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37] such as monotonicity, continuity, and differentiability. Recently, many generalizations and extensions have been made for the convexity, for example, s-convexity [38], strong convexity [39,40,41], preinvexity [42], GA-convexity [43], GG-convexity [44], Schur convexity [45,46,47,48,49], and others [50,51,52,53,54]. In particular, many remarkable inequalities can be found in the literature [55,56,57,58,59,60,61,62,63,64,65,66,67] via the convexity theory.
Chen [68] generalized the convex function to the s-convex function, gave the relation between the convex and s-convex functions, and established Jensen’s inequality for s-convex function as follows.
Let K be a convex subset of a real linear space and \(s\in (0, \infty )\) be a fixed real positive number. Then the mapping \(f: K\rightarrow \mathbb{R}\) is called s-convex on K if
for all \(x, y\in \mathbb{K}\) and \(\alpha , \beta \geq 0\) with \(\alpha +\beta \).
Lemma 1.1
([68])
Let \(\psi :I \rightarrow \mathbb{R}\) be a convex function defined on interval I. Then the following statements are true:
-
(i)
If ψ is non-negative, then ψ is s-convex for \(s\in (0, 1]\).
-
(ii)
If ψ is non-positive, then ψ is s-convex for \(s\in [1, \infty )\).
Theorem 1.2
([68])
Let \(i\in \{1,2,\ldots,n\}\), \(\alpha _{i}\geq 0\), \(Q_{n}=\sum_{i=1}^{n} \alpha _{i}^{\frac{1}{s}}>0\), and \(\psi :I\rightarrow \mathbb{R}\) be an s-convex function. Then
for all \(\xi _{i}\in I\).
2 Information divergence measures
Divergence measure is actually the distance between two probability distributions. Divergence measures have been introduced in the effort to solve the problems related to probability theory. Divergence measures have vast applications in a variety of fields such as economics, biology, signal processing, pattern recognition, computational learning, color image segmentation, magnetic resonance image analysis, and so on.
A class of information divergence measures, which is one of the important divergence measures due to its compact behavior, is the Csiszár ϕ-divergence [69] given below:
where \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\dots ,\eta _{n})\), \(\boldsymbol{\zeta } =(\zeta _{1},\zeta _{2},\dots ,\zeta _{n})\) are positive real n-tuples.
The Csiszár ϕ-divergence is a generalized measure of information on the convex function \(\phi : \mathbb{R}^{+}\rightarrow \mathbb{R}\), where the convexity ensures the non-negativity of divergence measures \({I}_{\phi }(\boldsymbol{\eta }, \boldsymbol{\zeta })\). The following Theorems 2.1 and 2.2 can be found in the literature [70, 71].
Theorem 2.1
If \(\phi :[0, \infty )\rightarrow \mathbb{R}\) is convex, then \({I}_{\phi }(\boldsymbol{\eta },\boldsymbol{\zeta })\) is jointly convex in η and ζ.
Theorem 2.2
Let \(\phi : \mathbb{R}^{+}\rightarrow \mathbb{R}^{+}\) be convex. Then, for every \(p, q \in \mathbb{R}^{n}_{+}\) with \(Q_{n}=\sum_{i=1} ^{n}\zeta _{i}\), we have
If ϕ is strictly convex, then equality holds in (2.1) if and only if
Corollary 2.3
Let \(\phi : \mathbb{R}^{+}\rightarrow \mathbb{R}^{+}\) be convex and normalized (\(\phi (1)=0\)) with \(\sum_{i=1}^{n}\eta _{i}=\sum_{i=1}^{n}\zeta _{i}\). Then we have
Equality holds in (2.2) if ϕ is strictly convex and \(\sum_{i=1}^{n}\eta _{i}=\sum_{i=1}^{n}\zeta _{i}\).
Many well-known distance functions or divergences can be obtained for a suitable choice of function ϕ, and they are frequently used in mathematical statistics, signal processing, and information theory. Some of the divergences are Kullback–Leibler, Renyi, Hellinger, Chi-square, Jeffery’s divergences, variational distance, and so on. Some brief introduction to these divergences is given below.
In probability and statistics, observed data is approximated by probability distribution. This approximation results in loss of information. The primitive object of information theory is to estimate how much information is in the data. Entropy is used to measure this information. Approximating a distribution by \(\boldsymbol{\zeta} (\boldsymbol{x})\) for which the actual distribution is \(\boldsymbol{\eta} (\boldsymbol{x})\) results in loss of information. KL-divergence, although not a true metric, is a useful measure of distance between the two distributions. The KL-divergence measure is the insufficiency of encoding the data with respect to the distribution ζ, rather than the true distribution η. The formula for KL-divergence can be obtained by choosing \(\phi (t)=t \log t\) in Csiszár divergence
The KL-divergence is non-negative if and only if \(\boldsymbol{\eta }= \boldsymbol{\zeta }\). However, it is not true distance between distributions, since it is not symmetric and does not satisfy the triangle inequality.
A logical alternative divergence or extension to KL-divergence is Jaffery’s divergence. It is the sum of the KL-divergence in both directions. It is defined by
which corresponds to ϕ-divergence for ϕ defined by
It exhibits the two properties of metric like KL-divergence but is also symmetric; however, it does not obey the triangle inequality. Its uses are similar to those of KL-divergence.
The Bhattacharyya divergence is defined by
which corresponds to ϕ-divergence for ϕ defined by
It satisfies the first three properties of metric but does not obey the triangle inequality. A nice feature of Bhattacharyya divergence is its limited range. Indeed its range is limited to make it quite attractive for a distance comparison.
The Bhattacharyya divergence is related to Hellinger divergence
corresponding to a ϕ-divergence for ϕ defined by
Hellinger divergence is in fact a proper metric because it satisfies non-negativity, symmetry, and triangle inequality properties. This makes it an ideal candidate for estimation and classification problems. Test statistics based on Hellinger divergence were developed for the independent samples drawn from two different continuous populations with a common parameter. It is used as a splitting criterion in decision trees, which is an effective way to address the imbalanced data problems. Hellinger divergence has deep roots in information theory and machine learning. It is extensively used in data analysis, especially when the objects being compared are high dimensional empirical probability distribution built from data.
Another ϕ-divergence is the total variational distance. The total variational distance is a distance measure for probability distribution, sometimes called statistical distance or variational distance, and it is defined by
which corresponds to a ϕ-divergence for ϕ defined by
Variational distance is a fundamental quantity in statistics and probability which appeared in many diverse applications. In information theory it is used to define strong typicality and asymptotic equipartition of sequences generated by sampling from a given distribution. In decision problems it arises naturally when discriminating the results of observation of two statistical hypotheses. In studying the ergodicity of Markov chains, it is used to define Dobrushin coefficient and establish the contraction property of transition probability distributions. Moreover, distance in total variation of probability measure is related via upper and lower bounds to an anthology of distance and distance metrics.
Another divergence measure is the Renyi divergence defined as
which corresponds to a ϕ-divergence for ϕ defined by
where \(\alpha >1\). Renyi divergence is related to Renyi entropy much like KL-divergence is related to Shannon’s entropy.
Some other important divergences can be obtained from Csiszár divergence which are given below.
Chi-square divergence. For \(\phi (z)=(z-1)^{2}\) (\(z>0\)) in ϕ-divergence. The \(\chi ^{2}\)-divergence is given by
and \(\chi ^{2}(\boldsymbol{\eta },\boldsymbol{\zeta })+\chi ^{2}( \boldsymbol{\zeta },\boldsymbol{\eta })\) is known as symmetric Chi- square divergence.
Triangular discrimination. For \(\phi (z)= \frac{(z-1)^{2}}{z+1}\) (\(z>0\)), the triangular discrimination is given by
Relative arithmetic-geometric divergence. For \(\phi (z)= \frac{z+1}{2}\log \frac{1+z}{2z}\) (\(z>0\)), the relative arithmetic-geometric divergence is given by
3 Inequalities for Csiszár divergence
Theorem 3.1
Let \(\phi :\mathbb{R}^{+}\rightarrow \mathbb{R}\) be an s-convex function, \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, and \(Q_{n}=\sum_{i=1}^{n}\zeta _{i} ^{\frac{1}{s}}\). Then one has
Proof
By taking \(\alpha _{i}\rightarrow \zeta _{i}\) and \(\xi _{i}\rightarrow \frac{ \eta _{i}}{\zeta _{i}}\) in Theorem 1.2, we get
which is equivalent to (3.1). □
Theorem 3.2
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, and \(Q_{n}=\sum_{i=1}^{n}\zeta _{i} ^{\frac{1}{s}}\). Then the following statements are true:
-
(i)
If \(\eta _{i}\geq \zeta _{i}\) for \(i\in \{1,2,\ldots,n \}\) and \(s\in (0, 1]\), then
$$ K(\boldsymbol{\eta },\boldsymbol{\zeta })\geq Q_{n}^{s}\frac{\sum_{i=1} ^{n}\zeta _{i}^{\frac{1-s}{s}}\eta _{i}}{\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}}\log \biggl(\frac{\sum_{i=1}^{n} \zeta _{i}^{ \frac{1-s}{s}}\eta _{i}}{\sum_{i=1}^{n}\zeta _{i}^{\frac{1}{s}}} \biggr). $$(3.2) -
(ii)
If \(\eta _{i}<\zeta _{i}\) for \(i\in \{1,2,\ldots,n\}\) and \(s\in [1, \infty )\), then inequality (3.2) holds.
Proof
(i) If \(\phi (z)=z\log z\), where \(z>0\), then \(\phi ^{\prime \prime }(z)= \frac{1}{z}\geq 0\), so \(\phi (z)\) is convex on \((0, \infty )\). Moreover, if \(z\geq 1\), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s\in (0, 1]\). Using \(\phi (z)=z\log z\) in Theorem 3.1, we get
which is equivalent to (3.2).
(ii) If \(z\leq 1\), then \(\phi (z)\leq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s\in [1, \infty )\); therefore, by utilizing Theorem 3.1, we obtain (3.3). □
Theorem 3.3
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s \in (0, 1]\). Then
Proof
If \(\phi (z)=(1-\sqrt{z})^{2}\), where \(z>0\), then \(\phi ^{\prime \prime }(z)= \frac{1}{2z}- \frac{\sqrt{z}-1}{2z^{\frac{3}{2}}}\geq 0\), so \(\phi (z)\) is convex on \((0, \infty )\). Moreover, if \(z>0\), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\). Using \(\phi (z)\) in Theorem 3.1, we have
which is equivalent to (3.4). □
Theorem 3.4
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s\in (0,1]\). Then
Proof
If \(\phi (z)=(z-1)^{2}\), where \(z>0\), then \(\phi ^{\prime \prime }(z)=2>0\), so \(\phi (z)\) is convex on \((0, \infty )\). Also, if \(z>0 \), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\). Utilizing \(\phi (z)=(z-1)^{2}\) in Theorem 3.1, we have
which is equivalent to (3.5). □
Theorem 3.5
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, and \(Q_{n}=\sum_{i=1}^{n}\zeta _{i} ^{\frac{1}{s}}\). Then the following statements are true:
-
(i)
If \(\eta _{i}\geq \zeta _{i}\) for \(i\in \{1,2,\ldots,n \}\) and \(s \in [1, \infty )\), then
$$ K(\boldsymbol{\zeta }, \boldsymbol{\eta })\geq Q_{n}^{s}\log \biggl(\frac{ \sum_{i=1}^{n}\zeta _{i}^{\frac{1}{s}}}{\sum_{i=1}^{n} \zeta _{i}^{ \frac{1-s}{s}}\eta _{i}} \biggr). $$(3.6) -
(ii)
If \(\eta _{i}<\zeta _{i}\) for \(i\in \{1,2,\ldots,n\}\) and \(s \in (0, 1]\), then inequality (3.6) holds.
Proof
(i) Let \(\phi (z)=-\log {z}\) (\(z>0\)). Then \(\phi ^{\prime \prime }(z)=\frac{1}{z ^{2}}>0\), so \(\phi (z)\) is convex on \((0, \infty )\). Moreover, if \(z\geq 1\), then \(\phi (z)\leq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s\in [1, \infty )\). Using \(\phi (z)=- \log {z}\) in Theorem 3.1, we get
which is equivalent to (3.6).
(ii) If \(z\leq 1\), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s\in (0, 1]\).
Similarly as above, using the function \(\phi (z)=-\log (z)\) in Theorem 3.1, we obtain (3.6). □
Theorem 3.6
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s \in (0, 1]\). Then
Proof
If \(\phi (z)=(z-1)\log z\) (\(z>0\)), then \(\phi ^{\prime \prime }(z)=\frac{z+1}{z ^{2}}\), so \(\phi (z)\) is convex on \((0, \infty )\). Moreover, if \(z>0\), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\). Using \(\phi (z)=(z-1)\log z\) in Theorem 3.1, we have
which is equivalent to (3.7). □
Theorem 3.7
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s \in (0, 1]\). Then
Proof
For \(\alpha >1\), the function \(\phi (z)=z^{\alpha }\) (\(z>0\)) is non-negative and convex. Therefore, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\). Using \(\phi (z)=z^{\alpha }\) in Theorem 3.1, we get
which is equivalent to (3.8). □
Theorem 3.8
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s \in (0, 1]\). Then
Proof
If \(\phi (z)=|z-1|\) (\(z\in \mathbb{R}\)), then clearly \(\phi (z)\) is convex on \(\mathbb{R}\). Moreover, for \(z\in \mathbb{R}\), \(\phi (z) \geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\). Using \(\phi (z)=|z-1|\) in Theorem 3.1, we get
which is equivalent to (3.9). □
Theorem 3.9
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s \in (0, 1]\). Then
Proof
If \(\phi (z)=(z-1)^{2}\) (\(z>0\)), then \(\phi ^{\prime \prime }(z)=2>0\), so \(\phi (z)\) is convex on \((0, \infty )\). Also, if \(z>0 \), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\).
From Theorem 3.4, we have
By interchanging \(\eta _{i}\) and \(\zeta _{i}\) in Theorem 3.4, we get
Adding (3.11) and (3.12), we get
which is equivalent to (3.10). □
Theorem 3.10
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s \in (0, 1]\). Then
Proof
If \(\phi (z)=\frac{(z-1)^{2}}{z+1}\) (\(z>0\)), then \(\phi ^{\prime \prime }(z)= \frac{8}{(z+1)^{3}}\geq 0\), so \(\phi (z)\) is convex on \((0, \infty )\). Moreover, if \(z>0\), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\). Using \(\phi (z)=\frac{(z-1)^{2}}{z+1}\) in Theorem 3.1, we have
which is equivalent to (3.13). □
Theorem 3.11
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, and \(Q_{n}=\sum_{i=1}^{n}\zeta _{i} ^{\frac{1}{s}}\). Then the following statements are true:
-
(i)
If \(\eta _{i}\geq \zeta _{i}\) for \(i\in \{1,2,\ldots,n \}\) and \(s \in [1, \infty )\), then
$$ G(\boldsymbol{\eta }, \boldsymbol{\zeta })\geq \frac{\sum_{i=1}^{n} \zeta _{i}^{\frac{1-s}{s}}+\sum_{i=1}^{n}\zeta _{i}^{\frac{1}{s}}}{2} \log \frac{\sum_{i=1}^{n}\zeta _{i}^{\frac{1}{s}}+\sum_{i=1}^{n}\zeta _{i}^{\frac{1-s}{s}}}{2\sum_{i=1}^{n}\zeta _{i}^{\frac{1-s}{s}}\eta _{i}}. $$(3.14) -
(ii)
If \(\eta _{i}<\zeta _{i}\) for \(i\in \{1,2,\ldots,n\}\) and \(s \in (0, 1] \), then inequality (3.14) holds.
Proof
(i) If \(\phi (z)=\frac{z+1}{2}\log \frac{1+z}{2z}\) (\(z>0\)), then \(\phi ^{\prime \prime }(z)=\frac{1}{2z^{2}(z+1)}>0\), so \(\phi (z)\) is convex on \((0, \infty )\). Moreover, if \(z\geq 1\), then \(\phi (z) \leq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s\in [1, \infty )\). Using \(\phi (z)\) in Theorem 3.1, we have
which is equivalent to (3.14).
(ii) If \(z\in (0,1]\), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s\in (0, 1]\). Similar to part (i), using Theorem 3.1, we obtain (3.14). □
Theorem 3.12
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(s \in (0, 1]\). Then
Proof
If \(\phi (z)=\frac{z+1}{2}\log \frac{1+z}{2z}\) (\(z>0\)). Then \(\phi ^{\prime \prime }(z)=\frac{1}{2z^{2}(z+1)}>0\), so \(\phi (z)\) is convex on \((0, \infty )\). Moreover, if \(z>0\), then \(\phi (z)\geq 0\). Hence, by Lemma 1.1, \(\phi (z)\) is s-convex for \(s \in (0, 1]\). From Theorem 3.11 we have
By interchanging \(\eta _{i}\) and \(\zeta _{i}\) in Theorem 3.11, we get
Adding (3.15) and (3.16), we obtain
namely
□
In the following theorem, we obtain a bound for Bhattacharyya divergence by utilizing an s-convex function that is not convex.
Theorem 3.13
Let \(\boldsymbol{\eta }=(\eta _{1},\eta _{2},\ldots,\eta _{n})\) and \(\boldsymbol{\zeta }=(\zeta _{1},\zeta _{2},\ldots,\zeta _{n})\) be two positive real n-tuples, \(Q_{n}=\sum_{i=1}^{n}\zeta _{i}^{ \frac{1}{s}}\) and \(0< s\leq \frac{1}{2}\). Then
Proof
First we show that \(\phi (z)=\sqrt{z}\) is s-convex for \(z>0\) and \(s\in (0, 1/2]\), namely we show that
for \(\lambda \in (0, 1)\) and \(s\in (0, 1/2]\).
Squaring both sides, we get
which implies that
Let \(\lambda =1/p\) (\(p>1\)). Then
for \(s\in (0, 1/2]\).
Namely,
for \(s\in (0, 1/2]\).
As \(\lambda \in (0, 1)\), \(1- \lambda \in (0, 1)\) and from (3.19), we have
From (3.19) and (3.20) we get (3.18), namely \(\phi (z)\) is s-convex for \(s \in (0,\frac{1}{2} ]\).
Now, using \(\phi (z)=\sqrt{z}\) in Theorem 3.1, we obtain
which is equivalent to (3.17). □
4 Conclusion
In the literature, there are several results for Jensen’s inequality by using convex functions. Particularly, there are many applications of Jensen’s inequality for convex functions in information theory. In this paper, we associated the results for s-convex functions with several divergences and proposed several applications of Jensen’s inequality for s-convex functions in information theory. We have obtained generalized inequalities for different divergences by using Jensen’s inequality for s-convex functions. The results obtained in this paper may also open the new door to obtaining other results in information theory for s-convex functions.
References
Pečarić, J.E., Proschan, F., Tong, Y.L.: Convex Functions, Partial Orderings, and Statistical Applications. Academic Press, Boston (1992)
Udrişte, C.: Convex Functions and Optimization Methods on Riemannian Manifolds. Kluwer Academic, Dordrecht (1994)
Huang, C.-X., Yang, Z.-C., Yi, T.-S., Zou, X.-F.: On the basins of attraction for a class of delay differential equations with non-monotone bistable nonlinearities. J. Differ. Equ. 256(7), 2101–2114 (2014)
Duan, L., Huang, C.-X.: Existence and global attractivity of almost periodic solutions for a delayed differential neoclassical growth model. Math. Methods Appl. Sci. 40(3), 814–822 (2017)
Duan, L., Huang, L.-H., Guo, Z.-Y., Fang, X.-W.: Periodic attractor for reaction-diffusion high-order Hopfield neural networks with time-varying delays. Comput. Math. Appl. 73(2), 233–245 (2017)
Wang, W.-S., Chen, Y.-Z.: Fast numerical valuation of options with jump under Merton’s model. J. Comput. Appl. Math. 318, 79–92 (2017)
Hu, H.-J., Liu, L.-Z.: Weighted inequalities for a general commutator associated to a singular integral operator satisfying a variant of Hörmander’s condition. Math. Notes 101(5–6), 830–840 (2017)
Cai, Z.-W., Huang, J.-H., Huang, L.-H.: Generalized Lyapunov–Razumikhin method for retarded differential inclusions: applications to discontinuous neural networks. Discrete Contin. Dyn. Syst. 22B(9), 3591–3614 (2017)
Hu, H.-J., Zou, X.-F.: Existence of an extinction wave in the Fisher equation with a shifting habitat. Proc. Am. Math. Soc. 145(11), 4763–4771 (2017)
Yang, C., Huang, L.-H.: New criteria on exponential synchronization and existence of periodic solutions of complex BAM networks with delays. J. Nonlinear Sci. Appl. 10(10), 5464–5482 (2017)
Tan, Y.-X., Huang, C.-X., Sun, B., Wang, T.: Dynamics of a class of delayed reaction-diffusion systems with Neumann boundary condition. J. Math. Anal. Appl. 458(2), 1115–1130 (2018)
Tang, W.-S., Zhang, J.-J.: Symplecticity-preserving continuous-stage Runge–Kutta–Nyström methods. Appl. Math. Comput. 323, 204–219 (2018)
Duan, L., Fang, X.-W., Huang, C.-X.: Global exponential convergence in a delayed almost periodic Nicholson’s blowflies model with discontinuous harvesting. Math. Methods Appl. Sci. 41(5), 1954–1965 (2018)
Liu, Z.-Y., Wu, N.-C., Qin, X.-R., Zhang, Y.-L.: Trigonometric transform splitting methods for real symmetric Toeplitz systems. Comput. Math. Appl. 75(8), 2782–2794 (2018)
Liu, B.-W., Tian, X.-M., Yang, L.-S., Huang, C.-X.: Periodic solutions for a Nicholson’s blowflies model with nonlinear mortality and continuously distributed delays. Acta Math. Appl. Sin. 41(1), 98–109 (2018)
Zhu, K.-X., Xie, Y.-Q., Zhou, F.: Pullback attractors for a damped semilinear wave equation with delays. Acta Math. Sin. 34(7), 1131–1150 (2018)
Zhang, Y.: On products of consecutive arithmetic progressions II. Acta Math. Hung. 156(1), 240–254 (2018)
Wang, J.-F., Chen, X.-Y., Huang, L.-H.: The number and stability of limit cycles for planar piecewise linear systems of node-saddle type. J. Math. Anal. Appl. 469(1), 405–427 (2019)
Li, J., Ying, J.-Y., Xie, D.-X.: On the analysis and application of an ion size-modified Poisson–Boltzmann equation. Nonlinear Anal., Real World Appl. 47, 188–203 (2019)
Jiang, Y.-J., Xu, X.-J.: A monotone finite volume method for time fractional Fokker–Planck equations. Sci. China Math. 62(4), 783–794 (2019)
Lin, L., Liu, Z.-Y.: An alternating projected gradient algorithm for nonnegative matrix factorization. Appl. Math. Comput. 217(24), 9997–10002 (2011)
Liu, Z.-Y., Zhang, Y.-L., Santos, J., Ralha, R.: On computing complex square roots of real matrices. Appl. Math. Lett. 25(10), 1565–1568 (2012)
Wang, W.-S.: High order stable Runge–Kutta methods for nonlinear generalized pantograph equations on the geometric mesh. Appl. Math. Model. 39(1), 270–283 (2015)
Li, J., Liu, F., Fang, L., Turner, I.: A novel finite volume method for the Riesz space distributed-order advection-diffusion equation. Appl. Math. Model. 46, 536–553 (2017)
Tan, Y.-X., Jing, K.: Existence and global exponential stability of almost periodic solution for delayed competitive neural networks with discontinuous activations. Math. Methods Appl. Sci. 39, 2821–2839 (2016)
Li, J.-L., Sun, G.-Y., Zhang, R.-M.: The numerical solution of scattering by infinite rough interfaces based on the integral equation method. Comput. Math. Appl. 71(7), 1491–1502 (2016)
Dai, Z.-F.: Comments on a new class of nonlinear conjugate gradient coefficients with global convergence properties. Appl. Math. Comput. 276, 297–300 (2016)
Dai, Z.-F., Chen, X.-H., Wen, F.-H.: A modified Perry’s conjugate gradient method-based derivative-free method for solving large-scale nonlinear monotone equations. Appl. Math. Comput. 270, 378–386 (2015)
Xie, D.-X., Li, J.: A new analysis of electrostatic free energy minimization and Poisson–Boltzmann equation for protein in ionic solvent. Nonlinear Anal., Real World Appl. 21, 185–196 (2015)
Tang, W.-S., Sun, Y.-J.: Construction of Runge–Kutta type methods for solving ordinary differential equations. Appl. Math. Comput. 234, 179–191 (2014)
Liu, Y.-C., Wu, J.: Fixed point theorems in piecewise continuous function spaces and applications to some nonlinear problems. Math. Methods Appl. Sci. 37(4), 508–517 (2014)
Li, X.-F., Tang, G.-J., Tang, B.-Q.: Stress field around a strike-slip fault in orthotropic elastic layers via a hypersingular integral equation. Comput. Math. Appl. 66(11), 2317–2326 (2013)
Jiang, Y.-J., Ma, J.-T.: Spectral collocation methods for Volterra-integro differential equations with noncompact kernels. J. Comput. Appl. Math. 244, 115–124 (2013)
Dai, Z.-F.: Two modified HS type conjugate gradient methods for unconstrained optimization problems. Nonlinear Anal. 74(3), 927–936 (2011)
Yang, X.-S., Zhu, Q.-X., Huang, C.-X.: Generalized lag-synchronization of chaotic mix-delayed systems with uncertain parameters and unknown perturbations. Nonlinear Anal., Real World Appl. 12(1), 93–105 (2011)
Zhou, W.-J., Zhang, L.: Global convergence of a regularized factorized quasi-Newton method for nonlinear least squares problems. Comput. Appl. Math. 29(2), 195–204 (2010)
Shi, H.-P., Zhang, H.-Q.: Existence of gap solitons in periodic discrete nonlinear Schrödinger equations. J. Math. Anal. Appl. 361(2), 411–419 (2010)
Adil Khan, M., Chu, Y.-M., Khan, T.U., Khan, J.: Some new inequalities of Hermite–Hadamard type for s-convex functions with applications. Open Math. 15(1), 1414–1430 (2017)
Song, Y.-Q., Adil Khan, M., Zaheer Ullah, S., Chu, Y.-M.: Integral inequalities involving strongly convex function. J. Funct. Spaces 2018, Article ID 6595921 (2018)
Zaheer Ullah, S., Adil Khan, M., Chu, Y.-M.: Majorization theorems for strongly convex functions. J. Inequal. Appl. 2019, Article ID 58 (2019)
Zaheer Ullah, S., Adil Khan, M., Khan, Z.A., Chu, Y.-M.: Integral majorization type inequalities for the functions in the sense of strong convexity. J. Funct. Spaces 2019, Article ID 9487823 (2019)
Khurshid, Y., Adil Khan, M., Chu, Y.-M., Khan, Z.A.: Hermite–Hadamard–Fejér inequalities for conformable fractional integrals via preinvex functions. J. Funct. Spaces 2019, Article ID 3146210 (2019)
Zhang, X.-M., Chu, Y.-M., Zhang, X.-H.: The Hermite–Hadamard type inequality of GA-convex functions and its applications. J. Inequal. Appl. 2010, Article ID 507560 (2010)
Khurshid, Y., Adil Khan, M., Chu, Y.-M.: Conformable integral inequalities of the Hermite–Hadamard type in terms of GG- and GA-convexities. J. Funct. Spaces 2019, Article ID 6926107 (2019)
Chu, Y.-M., Xia, W.-F., Zhao, T.-H.: Schur convexity for a class of symmetric functions. Sci. China Math. 53(2), 465–474 (2010)
Chu, Y.-M., Wang, G.-D., Zhang, X.-H.: Schur convexity and Hadamard’s inequality. Math. Inequal. Appl. 13(4), 725–731 (2010)
Chu, Y.-M., Wang, G.-D., Zhang, X.-H.: The Schur multiplicative and harmonic convexities of the complete symmetric function. Math. Nachr. 284(5–6), 653–663 (2011)
Chu, Y.-M., Xia, W.-F., Zhang, X.-H.: The Schur concavity, Schur multiplicative and harmonic convexities of the second dual form of the Hamy symmetric function with applications. J. Multivar. Anal. 105, 412–421 (2012)
Wu, S.-H., Chu, Y.-M.: Schur m-power convexity of generalized geometric Bonferroni mean involving three parameters. J. Inequal. Appl. 2019, Article ID 57 (2019)
Chu, Y.-M., Adil Khan, M., Ali, T., Dragomir, S.S.: Inequalities for α-fractional differentiable functions. J. Inequal. Appl. 2017, Article ID 93 (2017)
Adil Khan, M., Begum, S., Khurshid, Y., Chu, Y.-M.: Ostrowski type inequalities involving conformable fractional integrals. J. Inequal. Appl. 2018, Article ID 70 (2018)
Adil Khan, M., Chu, Y.-M., Kashuri, A., Liko, R., Ali, G.: Conformable fractional integrals versions of Hermite–Hadamard inequalities and their applications. J. Funct. Spaces 2018, Article ID 6928130 (2018)
Adil Khan, M., Khurshid, Y., Du, T.-S., Chu, Y.-M.: Generalization of Hermite–Hadamard type inequalities via conformable fractional integrals. J. Funct. Spaces 2018, Article ID 5357463 (2018)
Adil Khan, M., Wu, S.-H., Ullah, H., Chu, Y.-M.: Discrete majorization type inequalities for convex functions on rectangles. J. Inequal. Appl. 2019, Article ID 16 (2019)
Chu, Y.-M., Wang, M.-K.: Optimal Lehmer mean bounds for the Toader mean. Results Math. 61(3–4), 223–229 (2012)
Chu, Y.-M., Wang, M.-K., Qiu, S.-L.: Optimal combinations bounds of root-square and arithmetic means for Toader mean. Proc. Indian Acad. Sci. Math. Sci. 122(1), 41–51 (2012)
Yang, Z.-H., Qian, W.-M., Chu, Y.-M., Zhang, W.: Monotonicity rule for the quotient of two functions and its application. J. Inequal. Appl. 2017, Article ID 106 (2017)
Yang, Z.-H., Qian, W.-M., Chu, Y.-M., Zhang, W.: On rational bounds for the gamma function. J. Inequal. Appl. 2017, Article ID 210 (2017)
Qian, W.-M., Chu, Y.-M.: Sharp bounds for a special quasi-arithmetic mean in terms of arithmetic and geometric means with two parameters. J. Inequal. Appl. 2017, Article ID 374 (2017)
Huang, T.-R., Han, B.-W., Ma, X.-Y., Chu, Y.-M.: Optimal bounds for the generalized Euler–Mascheroni constant. J. Inequal. Appl. 2018, Article ID 118 (2018)
Huang, T.-R., Tan, S.-Y., Ma, X.-Y., Chu, Y.-M.: Monotonicity properties and bounds for the complete p-elliptic integrals. J. Inequal. Appl. 2018, Article ID 239 (2018)
Zhao, T.-H., Wang, M.-K., Zhang, W., Chu, Y.-M.: Quadratic transformation inequalities for Gaussian hypergeometric function. J. Inequal. Appl. 2018, Article ID 251 (2018)
Yang, Z.-H., Qian, W.-M., Chu, Y.-M.: Monotonicity properties and bounds involving the complete elliptic integrals of the first kind. Math. Inequal. Appl. 21(4), 1185–1199 (2018)
Yang, Z.-H., Chu, Y.-M., Zhang, W.: High accuracy asymptotic bounds for the complete elliptic integral of the second kind. Appl. Math. Comput. 348, 552–564 (2019)
Zhao, T.-H., Zhou, B.-C., Wang, M.-K., Chu, Y.-M.: On approximating the quasi-arithmetic mean. J. Inequal. Appl. 2019, Article ID 42 (2019)
Qiu, S.-L., Ma, X.-Y., Chu, Y.-M.: Sharp Landen transformation inequalities for hypergeometric functions, with applications. J. Math. Anal. Appl. 474(2), 1306–1337 (2019)
Wang, M.-K., Chu, Y.-M., Zhang, W.: Monotonicity and inequalities involving zero-balanced hypergeometric function. Math. Inequal. Appl. 22(2), 601–617 (2019)
Chen, X.-S.: New convex functions in linear spaces and Jensen’s discrete inequality. J. Inequal. Appl. 2013, Article ID 472 (2013)
Csiszár, I.: Information-type measures of difference of probability distributions and indirect observations. Studia Sci. Math. Hung. 2, 299–318 (1967)
Csiszár, I., Körner, J.: Information Theory. Academic Press, New York (1981)
Dragomir, S.S.: Some inequalities for the Csiszár ϕ-divergence when ϕ is and L-Lipschitzian function and applications. Ital. J. Pure Appl. Math. 15, 57–76 (2004)
Acknowledgements
The authors would like to express their sincere thanks to the editor and the anonymous reviewers for their helpful comments and suggestions.
Availability of data and materials
Not applicable.
Funding
The research was supported by the Natural Science Foundation of China (Grants Nos. 61673169, 61374086, 11371125, 11401191).
Author information
Authors and Affiliations
Contributions
All authors contributed equally to the writing of this paper. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Adil Khan, M., Hanif, M., Abdul Hameed Khan, Z. et al. Association of Jensen’s inequality for s-convex function with Csiszár divergence. J Inequal Appl 2019, 162 (2019). https://doi.org/10.1186/s13660-019-2112-9
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-019-2112-9