Research Article Bounds on Nonsymmetric Divergence Measure in terms of Other Symmetric and Nonsymmetric Divergence Measures

Similar documents
Research Article A Unified Weight Formula for Calculating the Sample Variance from Weighted Successive Differences

Research Article Some E-J Generalized Hausdorff Matrices Not of Type M

Review Article Incomplete Bivariate Fibonacci and Lucas p-polynomials

Research Article Nonexistence of Homoclinic Solutions for a Class of Discrete Hamiltonian Systems

Research Article A New Second-Order Iteration Method for Solving Nonlinear Equations

INEQUALITY FOR CONVEX FUNCTIONS. p i. q i

Research Article Approximate Riesz Algebra-Valued Derivations

Research Article Moment Inequality for ϕ-mixing Sequences and Its Applications

Comparison Study of Series Approximation. and Convergence between Chebyshev. and Legendre Series

Double Stage Shrinkage Estimator of Two Parameters. Generalized Exponential Distribution

Research Article A Note on Ergodicity of Systems with the Asymptotic Average Shadowing Property

Research Article Invariant Statistical Convergence of Sequences of Sets with respect to a Modulus Function

The Sampling Distribution of the Maximum. Likelihood Estimators for the Parameters of. Beta-Binomial Distribution

Research Article Quasiconvex Semidefinite Minimization Problem

Research Article Carleson Measure in Bergman-Orlicz Space of Polydisc

Research Article Generalized Residual Entropy and Upper Record Values

Research Article On q-bleimann, Butzer, and Hahn-Type Operators

New Inequalities of Hermite-Hadamard-like Type for Functions whose Second Derivatives in Absolute Value are Convex

Record Values from T-X Family of. Pareto-Exponential Distribution with. Properties and Simulations

Correspondence should be addressed to Wing-Sum Cheung,

Research Article Two Expanding Integrable Models of the Geng-Cao Hierarchy

Improved Class of Ratio -Cum- Product Estimators of Finite Population Mean in two Phase Sampling

Research Article Global Exponential Stability of Discrete-Time Multidirectional Associative Memory Neural Network with Variable Delays

ON POINTWISE BINOMIAL APPROXIMATION

Discrete Orthogonal Moment Features Using Chebyshev Polynomials

γn 1 (1 e γ } min min

Research Article Metric Divergence Measures and Information Value in Credit Scoring

Research Article An Alternative Estimator for Estimating the Finite Population Mean Using Auxiliary Information in Sample Surveys

Research Article Nonautonomous Discrete Neuron Model with Multiple Periodic and Eventually Periodic Solutions

On Orlicz N-frames. 1 Introduction. Renu Chugh 1,, Shashank Goel 2

A REFINEMENT OF JENSEN S INEQUALITY WITH APPLICATIONS. S. S. Dragomir 1. INTRODUCTION

Some Results on Certain Symmetric Circulant Matrices

Research Article Health Monitoring for a Structure Using Its Nonstationary Vibration

Lecture 7: October 18, 2017

Research Article The Arens Algebras of Vector-Valued Functions

Research Article On the Strong Convergence and Complete Convergence for Pairwise NQD Random Variables

Research Article Health Monitoring for a Structure Using Its Nonstationary Vibration

Self-normalized deviation inequalities with application to t-statistic

Uniform Strict Practical Stability Criteria for Impulsive Functional Differential Equations

Research Article Robust Linear Programming with Norm Uncertainty

Divergence criteria for improved selection rules

ECE 901 Lecture 13: Maximum Likelihood Estimation

10.6 ALTERNATING SERIES

Lecture 13: Maximum Likelihood Estimation

Research Article Powers of Complex Persymmetric Antitridiagonal Matrices with Constant Antidiagonals

Review Article Complete Convergence for Negatively Dependent Sequences of Random Variables

COMPLEX FACTORIZATIONS OF THE GENERALIZED FIBONACCI SEQUENCES {q n } Sang Pyo Jun

Finite Difference Approximation for Transport Equation with Shifts Arising in Neuronal Variability

On forward improvement iteration for stopping problems

INFINITE SEQUENCES AND SERIES

Dimension-free PAC-Bayesian bounds for the estimation of the mean of a random vector

SDS 321: Introduction to Probability and Statistics

Convergence of Random SP Iterative Scheme

NUMERICAL METHOD FOR SINGULARLY PERTURBED DELAY PARABOLIC PARTIAL DIFFERENTIAL EQUATIONS

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

POWER AKASH DISTRIBUTION AND ITS APPLICATION

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 3 9/11/2013. Large deviations Theory. Cramér s Theorem

Finite Difference Approximation for First- Order Hyperbolic Partial Differential Equation Arising in Neuronal Variability with Shifts

Testing Statistical Hypotheses with Fuzzy Data

Moment-entropy inequalities for a random vector

Bounds for the Positive nth-root of Positive Integers

Power Comparison of Some Goodness-of-fit Tests

Testing Statistical Hypotheses for Compare. Means with Vague Data

Some Exponential Ratio-Product Type Estimators using information on Auxiliary Attributes under Second Order Approximation

A RANK STATISTIC FOR NON-PARAMETRIC K-SAMPLE AND CHANGE POINT PROBLEMS

New Results for the Fibonacci Sequence Using Binet s Formula

Approximation by Superpositions of a Sigmoidal Function

A Note on the Symmetric Powers of the Standard Representation of S n

A New Measure of Divergence with its Application to Multi-Criteria Decision Making under Fuzzy Environment

DECOMPOSITION METHOD FOR SOLVING A SYSTEM OF THIRD-ORDER BOUNDARY VALUE PROBLEMS. Park Road, Islamabad, Pakistan

Convergence of random variables. (telegram style notes) P.J.C. Spreij

PAijpam.eu ON TENSOR PRODUCT DECOMPOSITION

Bayesian and E- Bayesian Method of Estimation of Parameter of Rayleigh Distribution- A Bayesian Approach under Linex Loss Function

Evapotranspiration Estimation Using Support Vector Machines and Hargreaves-Samani Equation for St. Johns, FL, USA

MONOTONICITY OF SEQUENCES INVOLVING GEOMETRIC MEANS OF POSITIVE SEQUENCES WITH LOGARITHMICAL CONVEXITY

IIT JAM Mathematical Statistics (MS) 2006 SECTION A

HAJEK-RENYI-TYPE INEQUALITY FOR SOME NONMONOTONIC FUNCTIONS OF ASSOCIATED RANDOM VARIABLES

6.3 Testing Series With Positive Terms

The Perturbation Bound for the Perron Vector of a Transition Probability Tensor

Control chart for number of customers in the system of M [X] / M / 1 Queueing system

A Study on Total Rebellion Number in Graphs

Average Number of Real Zeros of Random Fractional Polynomial-II

Unbiased Estimation. February 7-12, 2008

5.1 A mutual information bound based on metric entropy

Mathematical Modeling of Optimum 3 Step Stress Accelerated Life Testing for Generalized Pareto Distribution

The Choquet Integral with Respect to Fuzzy-Valued Set Functions

CHAPTER 4 BIVARIATE DISTRIBUTION EXTENSION

Optimization Results for a Generalized Coupon Collector Problem

Research Article On the Strong Laws for Weighted Sums of ρ -Mixing Random Variables

Goodness-Of-Fit For The Generalized Exponential Distribution. Abstract

A GRÜSS TYPE INEQUALITY FOR SEQUENCES OF VECTORS IN NORMED LINEAR SPACES AND APPLICATIONS

Varanasi , India. Corresponding author

Element sampling: Part 2

Estimation of the essential supremum of a regression function

Rates of Convergence for Quicksort

Confidence interval for the two-parameter exponentiated Gumbel distribution based on record values

AN OPEN-PLUS-CLOSED-LOOP APPROACH TO SYNCHRONIZATION OF CHAOTIC AND HYPERCHAOTIC MAPS

On an Application of Bayesian Estimation

A Family of Unbiased Estimators of Population Mean Using an Auxiliary Variable

Chapter 10: Power Series

Transcription:

Iteratioal Scholarly Research Notices Volume 04, Article ID 8037, 9 pages http://dx.doi.org/0./04/8037 Research Article Bouds o Nosymmetric Divergece Measure i terms of Other Symmetric ad Nosymmetric Divergece Measures K. C. Jai, ad Praphull Chhabra Departmet of Mathematics, Malaviya Natioal Istitute of Techology, Jaipur, Rajastha 3007, Idia B-, Staff Coloy, MNIT, Jaipur, Rajastha 3007, Idia Correspodece should be addressed to K. C. Jai; jaikc 003@yahoo.com Received Jue 04; Accepted September 04; Published 9 October 04 Academic Editor: Agelo De Satis Copyright 04 K. C. Jai ad P. Chhabra. This is a ope access article distributed uder the Creative Commos Attributio Licese, which permits urestricted use, distributio, ad reproductio i ay medium, provided the origial work is properly cited. Vajda (97)studiedageeralized divergece measureofcsiszar sclass, socalled Chi-m divergece measure. Variatioal distace ad Chi-square divergece are the special cases of this geeralized divergece measure at ad,respectively.i this work, oparametric osymmetric measure of divergece, a particular part of Vajda geeralized divergece at 4, is take ad characterized. Its bouds are studied i terms of some well-kow symmetric ad osymmetric divergece measures of Csiszar s class by usig well-kow iformatio iequalities. Compariso of this divergece with others is doe. Numerical illustratios (verificatio) regardig bouds of this divergece are preseted as well.. Itroductio Let Γ = {P = (p,p,p 3,...,p ) : p i > 0, p i = }, be the set of all complete fiite discrete probability distributios. If we take p i 0for some,,3,...,,the we have to suppose that 0f(0) = 0f(0/0) = 0. Csiszar [] itroduced a geeralized f-divergece measure, which is give by distributios P, Q Γ. These measures are as follows (see [ 7]): Triagular discrimiatio [] =Δ (p i ), p i + () C f f( p i ), () where f : (0, ) R (set of real umbers) is a real, cotiuous, ad covex fuctio ad P=(p,p,p 3,...,p ), Q=(q,q,q 3,...,q ) Γ,wherep i ad are probability massfuctios.maykowdivergecescabeobtaied from this geeralized measure by suitably defiig the covex fuctio f.someofthoseareasfollows... Symmetric Divergece Measures. Symmetric measures are those that are symmetric with respect to probability Helliger discrimiatio [3] =h(p, Q) JS divergece [4, ] = ( p i ), =I(P, Q) = [ p i log ( p i )+ q p i + log ( )], i p i + (Jai ad Mathur [6]) =P (p i ) 4 (p i + )(p i +q i ), p 3 i q3 i (3) (4) ()

Iteratioal Scholarly Research Notices (Jai ad Srivastava [7]) =E (p i ) 4. (6) 3/ (p i ).. Nosymmetric Divergece Measures. Nosymmetric measures are those that are ot symmetric with respect to probability distributios P, Q Γ. These measures are as follows (see [8 0]): Relative J-divergece [8] =J R (P, Q) = Relative iformatio [9] =K Relative AG divergece [0] =G(P, Q) = ( p i + ) log ( p i + ). p i (p i ) log ( p i + ), (7) p i log ( p i ), (8) We ca see that J R [F(Q, P) + G(Q, P)], Δ [ W(P, Q)], h B(P, Q), ad I (/)[F(P, Q) + F(Q, P)], wherew (p i /(p i + )) is the harmoic mea divergece, B p i is the geometric mea divergece, ad F p i log(p i /(p i + )) is the relative JS divergece []. Equatios (3), (4), ad (8) are also kow as Kolmogorov s measure, iformatio radius, ad directed divergece, respectively.. Nosymmetric Divergece Measure ad Its Properties I this sectio, we obtai osymmetric divergece measure for covex fuctio ad further defie the properties of fuctio ad divergece. Firstly, Theorem is well kow i literature []. Theorem. If the fuctio f is covex ad ormalized, that is, f() = 0, thec f (P, Q) ad its adjoit C f (Q, P) are both oegative ad covex i the pair of probability distributio (P, Q) Γ Γ. Now, let f : (0, ) R be a fuctio defied as f (t) =f (t) = (t )4 t 3, t (0, ), f () =0, (9) f (t) = (t )3 (t+3) t 4, (t ) (t) = t. (0) Properties of the fuctio defied by (0) are as follows. (a) Sice f () = 0, f (t) is a ormalized fuctio. (b) Sice (t) 0 for all t (0, ) f (t) is a covex fuctio as well. (c) Sice f (t) < 0 at (0, ) ad f (t) > 0 at (, ) f (t) is mootoically decreasig i (0, ) ad mootoically icreasig i (, ) ad f () = 0, () = 0. Now put f (t) i (); the followig ew divergece measure: χ 4 (Q, P) =C f (p i ) 4. () χ 4 (Q, P) is called adjoit of χ 4 ((p i ) 4 /q 3 i ) (after puttig cojugate covex fuctio of χ 4 (Q, P) i ()) ad it is a particular case of Chi-m divergece measure [] at m = 4, which is give by χ m ( p i m /q m i ), m. Properties of the divergece measure defied i () are as follows. (a) I view of Theorem, wecasaythatχ 4 (Q, P) is covex ad oegative i the pair of probability distributio (P, Q) Γ Γ. (b) χ 4 (Q, P) = 0 if P=Qor p i = (attaiig its miimum value). (c) Sice χ 4 (Q, P) =χ 4 (P,Q) χ 4 (Q, P) is a osymmetric divergece measure. 3. Iformatio Iequalities of Csiszar s Class I this sectio, we are takig well-kow iformatio iequalities o C f (P, Q), whichisgivebytheorem. Such iequalities are, for istace, eeded i order to calculate the relative efficiecy of two divergeces. This theorem is due to literature [], which relates two geeralizedf-divergece measures. Theorem. Let f,f : I R + R be two covex ad ormalized fuctios, that is, f () = f () = 0, adsuppose the followig assumptios. (a) f ad f are twice differetiable o (α, β) where 0< α β<, α =β. (b) There exist the real costats m, M such that m<m ad m f (t) M, () (t) where (t) > 0 for all t (α, β). p 3 i

Iteratioal Scholarly Research Notices 3 If P, Q Γ ad satisfyig the assumptio 0<α p i / β<, the oe has the followig iequalities: mc f (P, Q) C f (P, Q) MC f (P, Q), (3) where C f (P, Q) is give by (). 4. Bouds i terms of Symmetric Divergece Measures Nowithissectio,weobtaiboudsofdivergecemeasure () i terms of other symmetric divergece measures by usig Theorem. Propositio 3. Let Δ(P, Q) ad χ 4 (Q, P) be defied as i () ad (),respectively.forp,q Γ, oe has the followig. (a) If 0<α<,the 0 χ 4 (P, Q) 3 max [(α ) (α+) 3 α, (β ) (β + ) 3 β ]Δ(P, Q). (b) If α=,the 0 χ 4 (P, Q) 3(β ) (β + ) 3 Proof. Let us cosider (4) β Δ (P, Q). () f (t) = (t ), t (0, ), t+ f () =0, f (t) = (t )(t+3) (t+), (t) = 8 (t+) 3. (6) Sice (t) > 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f (p i ) p i + =Δ(P, Q). (7) Now, let g(t) = (t)/f (t) = 3(t ) (t + ) 3 /t,where (t) ad f (t) are give by (0)ad(6), respectively, ad g (t) = 3 (t )(t+) ( t) t 6, g (t) = 3(t4 6t 3 t + 0t + ) t 7. If g (t) = 0 t =, t=,adt=. (8) It is clear that g(t) is decreasig i (0, ) ad [, ) but icreasig i [, ). Also g(t) has a miimum ad maximum value at t= ad t=,respectively,becauseg () = 4 > 0 ad g () = 6/6 < 0,so Ad (a) if 0<α<,the if t (0, ) g (t) =g() =0. (9) M= sup g (t) = max {g (α), g (β)} = max [ 3(α ) (α+) 3 α, 3(β ) (β + ) 3 β ]; (b) if α=,the (0) M= sup g (t) =g(β)= 3(β ) (β + ) 3 t (,β) β. () Results (4)ad() are obtaied by usig (), (7), (9), (0), ad () i(3), after iterchagig P ad Q. Propositio 4. Let h(p, Q) ad χ 4 (Q, P) be defied as i (3) ad (),respectively.forp, Q Γ, oe has the followig. (a) If 0<α<,the 0 χ 4 (P, Q) 48max [ (α ) (β ), ]h(p, Q). α 7/ β 7/ (b) If α=,the 0 χ 4 (P, Q) Proof. Let us cosider () 48(β ) β 7/ h (P, Q). (3) ( t) f (t) =, t (0, ), f () =0, f (t) = (t / ), (t) = 4t 3/. (4) Sice (t) > 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f ( p i ) =h(p, Q). ()

4 Iteratioal Scholarly Research Notices Now, let g(t) = (t)/f (t) = 48(t ) /t 7/,where (t) ad (t) are give by (0)ad(4), respectively, ad g 4 (t )(7 3t) (t) =, t 9/ g (t) = (6) (t 70t + 63). t / If g (t) = 0 t = ad t=7/3. It is clear that g(t) is decreasig i (0, ) ad [7/3, ) but icreasig i [, 7/3). Also g(t) has a miimum ad maximum value at t= ad t = 7/3,respectively,becauseg () = 96 > 0 ad g (.33) = 9/93 < 0,so if g (t) =g() =0. (7) t (0, ) Ad (a) if 0<α<,the M= sup g (t) = max {g (α),g(β)} = max [ 48(α ) (8) 48(β ), ]; α 7/ β 7/ (b) if α=,the 48(β ) M= sup g (t) =g(β)=. t (,β) β 7/ (9) Results () ad(3) are obtaied by usig (), (), (7), (8), ad (9)i(3), after iterchagig P ad Q. Sice (t) > 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f [ p i log ( p i )+ q p i + log ( )] i p i + =I(P, Q). (33) Now, let g(t) = (t)/f (t) = 4(t ) (t + )/t 4,where (t) ad f (t) are give by (0)ad(3), respectively, ad g (t) = 4 (t ) ( t +t+4) t, g (t) =48( t 3 3 t 4 6 t + 0 t 6 ). (34) If g (t) = 0 t =, t = ( 7)/,adt = ( + 7)/. It is clear that g(t) is decreasig i (0, ) ad [( + 7)/, ) but icreasig i [, ( + 7)/). Also g(t) has a miimum ad maximum value at t= ad t = ( + 7)/, respectively,becauseg () = 96 > 0 ad g (( + 7)/) = 6/86 < 0,so Ad (a) if 0<α<,the if t (0, ) g (t) =g() =0. (3) Propositio. Let I(P, Q) ad χ 4 (Q, P) be defied as i (4) ad (),respectively.forp, Q Γ, oe has the followig. (a) If 0<α<,the 0 χ 4 (P, Q) M= sup g (t) = max {g (α), g (β)} = max [ 4(α ) (α+) α 4, 4(β ) (β + ) β 4 ]; (36) 4max [ (α ) (α+) α 4, (β ) (β + ) β 4 ]I(P, Q). (b) If α=,the (30) 0 χ 4 (P, Q) 4(β ) (β + ) β 4 I (P, Q). (3) Proof. Let us cosider f (t) = t log t+t+ f () =0, f (t) = t log ( t+ ), (t) = t (t+). log ( ), t (0, ), t+ (3) (b) if α=,the M= sup g (t) =g(β)= 4(β ) (β + ) t (,β) β 4. (37) Results (30)ad(3) are obtaied by usig (), (33), (3), (36), ad (37)i(3), after iterchagig P ad Q. Propositio 6. Let P (P, Q) ad χ 4 (Q, P) be defied as i () ad (),respectively. For P, Q Γ,oehas β +β 4 +β 3 +β +β+ P (P, Q) χ 4 (P, Q) α +α 4 +α 3 +α +α+ P (P, Q). (38)

Iteratioal Scholarly Research Notices Proof. Let us cosider f (t) = (t )4 (t+)(t +) t 3, t (0, ), f () =0, f (t) = (t )3 (4t 4 +3t 3 +3t +3t+3) t 4, (t ) (t) = t [t +6t 4 +6t 3 +6t + 6t + ]. (39) Sice (t) 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f (p i ) 4 (p i + )(p i +q i ) =P (P, Q). p 3 i q3 i (40) Now, let g(t) = (t)/f (t) = /(t +t 4 +t 3 +t +t+), where (t) ad f (t) are give by (0)ad(39), respectively, ad g (t) = (0t4 +4t 3 +3t +t+) (t +t 4 +t 3 +t +t+) <0. (4) It is clear that g(t) is always decreasig i (0, ),so if g (t) =g(β)= β +β 4 +β 3 +β +β+, M= sup g (t) =g(α) = α +α 4 +α 3 +α +α+. (4) Result (38) isobtaiedbyusig(), (40), ad (4) i(3), after iterchagig P ad Q. Propositio 7. Let E (P, Q) ad χ 4 (Q, P) be defied as i (6) ad (),respectively.forp, Q Γ,oehas 6 β 3/ (β +6β+) E (P, Q) χ 4 (P, Q) Proof. Let us cosider 6 α 3/ (α +6α+) E (P, Q). f (t) = (t )4, t (0, ), t 3/ f () =0, f (t) = (t )3 (t + 3) t /, (t) = 3(t ) (t +6t+) 4t 7/. (43) (44) Sice (t) 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f (p i ) 4 (p i ) 3/ =E (P, Q). (4) Now, let g(t) = (t)/f (t) = 6/t3/ (t +6t+),where (t) ad f (t) are give by (0)ad(44), respectively, ad g (t) = 40 (7t +6t+3) <0. (46) t / (t +6t+) It is clear that g(t) is always decreasig i (0, ),so if g (t) =g(β)= 6 β 3/ (β +6β+), 6 M= sup g (t) =g(α) = α 3/ (α +6α+). (47) Result (43) isobtaiedbyusig(), (4), ad (47) i(3), after iterchagig P ad Q.. Bouds i terms of Nosymmetric Divergece Measures Nowithissectio,weobtaiboudsofdivergecemeasure () i terms of other osymmetric divergece measures by usig Theorem. Propositio 8. Let J R (P, Q) ad χ 4 (Q, P) be defied as i (7) ad (),respectively.forp, Q Γ, oe has the followig. (a) If 0<α<,the 0 χ 4 (α ) (P, Q) max [ (α+3) α, (β ) (β + 3) β ] J R (Q, P). [ ] (48) (b) If α=,the Proof. Let us cosider 0 χ 4 (P, Q) (β ) (β + 3) β J R (Q, P). (49) f (t) = (t ) log ( t+ ), t (0, ), f () =0, f (t) = (t )+log (t+ t+ ), (t+3) (t) = (t+). (0)

6 Iteratioal Scholarly Research Notices Sice (t) > 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f (p i ) log ( p i + )=J R (P, Q). () Proof. Let us cosider f (t) =tlog t, t (0, ), f () =0, f (t) =+log t, (8) Now, let g(t) = (t)/f (t) = (t ) /(t + 3)t,where (t) ad f (t) are give by (0)ad(0), respectively, ad g (t) = (t )(t 3 +3t 6t ) (t+3) t 6, g (t) = 4 ( 3t6 9t +t 4 +90t 3 +87t 0t 3) (t+3) t 7. () If g (t) = 0 t =, t=,adt 6/9. It is clear that g(t) is decreasig i (0, ) ad [6/9, ) but icreasig i [, 6/9). Also g(t) has a miimum ad maximum value at t= ad t = 6/9, respectively,becauseg () = 4 > 0 ad g (6/9) 7/0 < 0,so Ad (a) if 0<α<,the if t (0, ) g (t) =g() =0. (3) M= sup g (t) = max {g (α),g(β)} (α ) = max [ (α+3) α, (β ) (β + 3) β ] ; [ ] (b) if α=,the (4) M= sup g (t) =g(β) = (β ) t (,β) (β + 3) β. () Results (48) ad(49) are obtaied by usig (), (), (3), (4), ad ()i(3), after iterchagig P ad Q. Propositio 9. Let K(P, Q) ad χ 4 (Q, P) be defied as i (8) ad (),respectively.forp, Q Γ, oe has the followig. (a) If 0<α<,the 0 χ 4 (P, Q) max [ (α ) (β ) α 4, β 4 ]K(Q, P). (b) If α=,the 0 χ 4 (P, Q) (6) (β ) β 4 K (Q, P). (7) (t) = t. Sice (t) > 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f p i log ( p i )=K(P, Q). (9) Now, let g(t) = (t)/f (t) = (t ) /t 4,where (t) ad (t) are give by (0)ad(8), respectively, ad g 4 (t )( t) (t) = t, g (t) =4( 3 t 4 t + 0 (60) t 6 ). If g (t) = 0 t = ad t=. It is clear that g(t) is decreasig i (0, ) ad [, ) but icreasig i [, ). Also g(t) has a miimum ad maximum value at t= ad t=,respectively,becauseg () = 4 > 0 ad g () = 3/4 < 0,so if g (t) =g() =0. (6) t (0, ) Ad (a) if 0<α<,the M= sup g (t) = max {g (α),g(β)} (6) = max [ (α ) (β ) α 4, β 4 ]; (b) if α=,the (β ) M= sup g (t) =g(β)= t (,β) β 4. (63) Results (6) ad(7) are obtaied by usig (), (9), (6), (6), ad (63) i(3), after iterchagig P ad Q. Propositio 0. Let G(P, Q) ad χ 4 (Q, P) be defied as i (9) ad (),respectively.forp, Q Γ, oe has the followig. (a) If 0<α<,the 0 χ 4 (P, Q) 4max [ (α ) (α+) α 3, (β ) (β + ) β 3 ]G(Q, P). (64)

Iteratioal Scholarly Research Notices 7 Table : (=0,p=/,q=/). x i 0 3 4 6 7 8 9 0 p(x i )=p i q(x i )= p i 04 74 0 69 74 0 69 4 04 47 8 3 8 38 983 97 36 0 73 46 97 83 63 6 73 46 89 84 0 68 46 300 97 8 68 6 040 97 4 04 337 449 667 7 93 7 6 04 63 6 (b) If α=,the 0 χ 4 (P, Q) 4(β ) (β + ) β 3 G (Q, P). (6) Proof. Let us cosider f (t) = t+ f () =0, log ( t+ ), t (0, ), t f (t) = [log (t+ t ) t ], (t) = t (t+). (66) Sice (t) > 0 for all t>0ad f () = 0, f (t) is a covex ad ormalized fuctio, respectively. Now put f (t) i (); C f ( p i + ) log ( p i + )=G(P, Q). (67) p i Now, let g(t) = (t)/f (t) = 4(t ) (t + )/t 3,where (t) ad f (t) are give by (0)ad(66), respectively, ad g 4 (t )(t+3) (t) = t 4, g (t) =4( t 3 6 t 4 + t ). (68) If g (t) = 0 t = ad t= 3. It is clear that g(t) is decreasig i (0, ) ad icreasig i [, ). Also g(t) has a miimum value at t=,becauseg () = 0 > 0,so Ad (a) if 0<α<,the M= sup g (t) = max {g (α), g (β)} if t (0, ) g (t) =g() =0. (69) = max [ 4(α ) (α+) α 3, 4(β ) (β + ) β 3 ]; (70) (b) if α=,the M= sup g (t) =g(β)= 4(β ) (β + ) t (,β) β 3. (7) Results (64) ad(6) are obtaied by usig (), (67), (69), (70), ad (7) i(3), after iterchagig P ad Q. 6. Numerical Illustratio I this sectio, we give two examples for calculatig the divergeces Δ(P, Q), h(p, Q), J R (Q, P), K(Q, P),adχ 4 (P, Q) ad verify iequalities (4), (), (48), ad (6) orverify bouds of χ 4 (P, Q). Example. Let P be the biomial probability distributio with parameters ( = 0, p = /) ad Q its approximated Poisso probability distributio with parameter (λ = p = ), for the radom variable X;the we have Table. By usig Table, the followig: α(= 6 ) p i β(= 89 84 ), Δ h J R (Q, P) = K (Q, P) = χ 4 (p i ) p i + 9, ( p i ) 3 0, ( p i ) log ( p i + p i ) 3 397, log ( ) 9 p i 70, (p i ) 4 q 3 i 9 7. (7) Put the approximated umerical values from (7) i(4), (), (48), ad (6) ad verify iequalities (4), (), (48), ad (6) for p=/.

8 Iteratioal Scholarly Research Notices Table : ( = 0, p = 7/0, q = 3/0). x i 0 3 4 6 7 8 9 0 p(x i )=p i q(x i )= p i 9049 0 0 3778 0 9 69 98 3 0 7 470 940 77 3 39 6 47 36 787 34 9 6 8 83 06 67 6 47 368 83 03 4 93 93 70 46 4 93 703 9 66 7 00 767 890 497 73 603 9 60 33 4 77 60 499 30 979 Example. Let P be the biomial probability distributio with parameters ( =0, p = 7/0) adq its approximated Poisso probability distributio with parameter (λ = p = 7), for the radom variable X;the we have Table. By usig Table,wegetthefollowig: α(= 77 ) p i β(= 703 9 ), Δ h J R (Q, P) = K (Q, P) = χ 4 (p i ) p i + 60 883, ( p i ) 498, ( p i ) log ( p i + p i ) 99 389, log ( ) 4 p i 77, (p i ) 4 q 3 i 69 94. (73) Put the approximated umerical values from (73) i(4), (), (48), ad (6)adverifyiequalities(4), (), (48), ad (6) for p = 7/0. Similarly, we ca verify iequalities (or verify bouds of χ 4 (P, Q))(30), (38), (43), ad (64). Figure shows the behavior of covex fuctio i (0, ). Fuctio is decreasig i (0, ) ad icreasig i (, ). Figure shows the behavior of χ 4 (P, Q), E (P, Q), K(P, Q), Δ(P, Q), adg(p, Q). Wehavecosideredp i = (a, a), = ( a, a), wherea (0, ). ItisclearfromFigure that the divergece χ 4 (P, Q) has a steeper slope tha E (P, Q), K(P, Q), Δ(P, Q),adG(P, Q). 0 8 6 4. 0.8 0.6 0.4 0. 0 4 6 8 0 Figure : Covex fuctio f (t). 0 0. Chi divergece () Divergece (6) Relative etropy a Triagular discrimiatio Relative AG divergece Figure : Compariso of divergeces. 7. Coclusios May research papers have bee studied by Taeja, Kumar, Dragomir, Jai, ad others, who gave the idea of divergece measures, their properties, their bouds, ad relatios with other measures. Taeja especially did a lot of quality work i this field: for istace, i [3] he derived bouds o differet osymmetric divergeces i terms of differet symmetric divergeces, i [4] he itroduced ew geeralized divergeces ad ew divergeces as a result of differece of meas ad characterized their properties ad bouds, ad i [] ew iequalities amog oegative differeces arisig from seve meas have bee itroduced ad correlatios with geeralized triagular discrimiatio ad some ew geeratig measures with their expoetial represetatios have also bee preseted.

Iteratioal Scholarly Research Notices 9 Divergece measures have bee demostrated to be very useful i a variety of disciplies such as athropology, geetics, fiace, ecoomics ad political sciece, biology, aalysis of cotigecy tables, approximatio of probability distributios, sigal processig, patter recogitio, sesor etworks, testig of the order i a Markov chai, risk for biary experimets, regio segmetatio, ad estimatio. This paper also defies the properties ad bouds of Vajda s divergece ad derives ew relatios with other symmetric ad osymmetric well-kow divergece measures. [4] I. J. Taeja, O symmetric ad o symmetric divergece measures ad their geeralizatios, Advaces i Imagig ad Electro Physics,vol.38,pp.77 0,00. [] I. J. Taeja, Seve meas, geeralized triagular discrimiatio, ad geeratig divergece measures, Iformatio, vol.4, o., pp. 98 39, 03. Coflict of Iterests The authors declare that there is o coflict of iterests regardig the publicatio of this paper. Refereces [] I. Csiszar, Iformatio type measures of differeces of probability distributio ad idirect observatios, Studia Scietiarum Mathematicarum Hugarica, vol., pp. 99 38, 967. [] D. Dacuha-Castelle, Ecole d Ete de Probabilites de Sait-Flour VII 977, Spriger, Berli, Germay, 978. [3] E. Helliger, Neue begrüdug der theorie der quadratische forme vo uedliche viele veräderliche, Joural für Die Reie ud Agewadte Mathematik,vol.36,pp.0 7,909. [4] J. Burbea ad C. R. Rao, O the covexity of some divergece measures based o etropy fuctios, IEEE Trasactios o Iformatio Theory,vol.8,o.3,pp.489 49,98. [] R. Sibso, Iformatio radius, Zeitschrift für Wahrscheilichkeitstheorie ud Verwadte Gebiete, vol.4,o.,pp.49 60, 969. [6] K. C. Jai ad R. Mathur, A symmetric divergece measure ad its bouds, Tamkag Joural of Mathematics, vol. 4, o. 4, pp. 493 03, 0. [7] K. C. Jai ad A. Srivastava, O symmetric iformatio divergece measures of Csiszar s f-divergece class, Joural of Applied Mathematics, Statistics ad Iformatics, vol.3,o., 007. [8] S. S. Dragomir, V. Gluscevic, ad C. E. M. Pearce, Approximatio for the Csiszar f-divergece via midpoit iequalities, i Iequality Theory ad Applicatios, Y.J.Cho,J.K.Kim,adS. S.Dragomir,Eds.,vol.,pp.39 4,NovaSciecePublishers, Hutigto, NY, USA, 00. [9] S. Kullback ad R. A. Leibler, O iformatio ad sufficiecy, Aals of Mathematical Statistics,vol.,pp.79 86,9. [0] I. J. Taeja, New developmets i geeralized iformatio measures, Advaces i Imagig ad Electro Physics,vol.9,pp. 37 3, 99. [] I. Vajda, O the f-divergece ad sigularity of probability measures, Periodica Mathematica Hugarica, vol., pp. 3 34, 97. [] I. J. Taeja, Geeralized symmetric divergece measures ad iequalities, RGMIA Research Report Collectio, vol.7,o.4, article 9, 004. [3] I. J. Taeja, Bouds o o-symmetric divergece measures i terms of symmetric divergece measures, Joural of Combiatorics, Iformatio & System Scieces, vol.9,o. 4,pp. 34, 00.

Advaces i Operatios Research http://www.hidawi.com Volume 04 Advaces i Decisio Scieces http://www.hidawi.com Volume 04 Joural of Applied Mathematics Algebra http://www.hidawi.com http://www.hidawi.com Volume 04 Joural of Probability ad Statistics Volume 04 The Scietific World Joural http://www.hidawi.com http://www.hidawi.com Volume 04 Iteratioal Joural of Differetial Equatios http://www.hidawi.com Volume 04 Volume 04 Submit your mauscripts at http://www.hidawi.com Iteratioal Joural of Advaces i Combiatorics http://www.hidawi.com Mathematical Physics http://www.hidawi.com Volume 04 Joural of Complex Aalysis http://www.hidawi.com Volume 04 Iteratioal Joural of Mathematics ad Mathematical Scieces Mathematical Problems i Egieerig Joural of Mathematics http://www.hidawi.com Volume 04 http://www.hidawi.com Volume 04 Volume 04 http://www.hidawi.com Volume 04 Discrete Mathematics Joural of Volume 04 http://www.hidawi.com Discrete Dyamics i Nature ad Society Joural of Fuctio Spaces http://www.hidawi.com Abstract ad Applied Aalysis Volume 04 http://www.hidawi.com Volume 04 http://www.hidawi.com Volume 04 Iteratioal Joural of Joural of Stochastic Aalysis Optimizatio http://www.hidawi.com http://www.hidawi.com Volume 04 Volume 04