1 Introduction

Let (G,) be an Abelian group. Let I denote the open unit interval (0,1). Let ℝ and ℂ denote the set of real and complex numbers, respectively. Let R + ={xRx>0} be a set of positive real numbers and R k ={xRx>k>0} for some kR.

Further, let

Γ n o = { P = ( p 1 , p 2 , , p n ) | 0 < p k < 1 , k = 1 n p k = 1 }

denote the set of all n-ary discrete complete probability distributions (without zero probabilities), that is, Γ n o is the class of discrete distributions on a finite set Ω of cardinality n with n2. Over the years, many distance measures between discrete probability distributions have been proposed. The Hellinger coefficient, the Jeffreys distance, the Chernoff coefficient, the directed divergence, and its symmetrization J-divergence are examples of such measures (see [1] and [2]).

Almost all similarity, affinity or distance measures μ n : Γ n o × Γ n o R + that have been proposed between two discrete probability distributions can be represented in the sum form

μ n (P,Q)= k = 1 n ϕ( p k , q k ),
(1.1)

where ϕ:I×IR is a real-valued function on unit square, or a monotonic transformation of the right side of (1.1), that is,

μ n (P,Q)=ψ ( k = 1 n ϕ ( p k , q k ) ) ,
(1.2)

where ψ:R R + is an increasing function on ℝ. The function ϕ is called a generating function. It is also referred to as the kernel of μ n (P,Q).

In information theory, for P and Q in Γ n o , the symmetric divergence of degree α is defined as

J n , α (P,Q)= 1 2 α 1 1 [ k = 1 n ( p k α q k 1 α + p k 1 α q k α ) 2 ] .

It is easy to see that J n , α (P,Q) is symmetric. That is, J n , α (P,Q)= J n , α (Q,P) for all P,Q Γ n o . Moreover, it satisfies the composition law

J n m , α ( P R , Q S ) + J n m , α ( P S , Q R ) = 2 J n , α ( P , Q ) + 2 J m , α ( R , S ) + λ J n , α ( P , Q ) J m , α ( R , S )

for all P,Q Γ n o and R,S Γ m o where λ= 2 α 1 1 and

PR=( p 1 r 1 , p 1 r 2 ,, p 1 r m , p 2 r 1 ,, p 2 r m ,, p n r m ).

In view of this, symmetrically compositive statistical distance measures are defined as follows. A sequence of symmetric measures { μ n } is said to be symmetrically compositive if for some λR,

μ n m ( P R , Q S ) + μ n m ( P S , Q R ) = 2 μ n ( P , Q ) + 2 μ m ( R , S ) + λ μ n ( P , Q ) μ m ( R , S )

for all P,Q Γ n o , S,R Γ m o , where

PR=( p 1 r 1 , p 1 r 2 ,, p 1 r m , p 2 r 1 ,, p 2 r m ,, p n r m ).

Chung, Kannappan, Ng and Sahoo [1] characterized symmetrically compositive sum-form distance measures with a measurable generating function. The following functional equation:

(FE) f(pr,qs)+f(ps,qr)=f(p,q)f(r,s)

holding for all p,q,r,sI was instrumental in the characterization of symmetrically compositive sum-form distance measures. They proved the following theorem giving the general solution of this functional equation (FE).

Suppose f: I 2 R satisfies the functional equation (FE), that is,

f(pr,qs)+f(ps,qr)=f(p,q)f(r,s)

for all p,q,r,sI. Then

f(p,q)= M 1 (p) M 2 (q)+ M 1 (q) M 2 (p),
(1.3)

where M 1 , M 2 :RC are multiplicative functions. Further, either M 1 and M 2 are both real or M 2 is the complex conjugate of M 1 . The converse is also true.

The stability of the functional equation (FE), as well as the four generalizations of (FE), namely,

(F E f g ) f(pr,qs)+f(ps,qr)=f(p,q)g(r,s),

(F E g f ) f(pr,qs)+f(ps,qr)=g(p,q)f(r,s),

(F E g g ) f(pr,qs)+f(ps,qr)=g(p,q)g(r,s),

(F E g h ) f(pr,qs)+f(ps,qr)=g(p,q)h(r,s)

for all p,q,r,sG, were studied by Kim and Sahoo in [3, 4]. For other functional equations similar to (FE), the interested reader should refer to [58], and [9].

The present work continues the study for the stability of the Pexider type functional equation of (FE) added a cocycle property to the conditions in the results [3, 4]. These functional equations arise in the characterization of symmetrically compositive sum-form distance measures, products of some multiplicative functions. In reduction, they can be represented as a (hyperbolic) cosine (sine, trigonometric) functional equation, exponential, and Jensen functional equation, respectively.

Tabor [10] investigated the cocycle property. The definition of cocycle as follows:

Definition 1 A function θ: G 2 R is a cocycle if it satisfies the equation

θ(a,bc)θ(b,c)=θ(ab,c)θ(a,b),a,b,cG.

For example, if F(x,y)= f ( x ) f ( y ) f ( x y ) for a function f:R R + , then F is a cocycle. Also if θ(x,y)=ln(x)ln(y) for a function θ: R + 2 (R,+), then θ is a cocycle, that is, θ(a,bc)+θ(b,c)=θ(ab,c)+θ(a,b), and in this case, it is well known that θ(x,y) is represented by B(x,y)+M(xy)M(x)M(y) where B is an arbitrary skew-symmetric biadditive function and M is some function [11]. If θ(x,y)= a ln ( x ) ln ( y ) , then θ: R + 2 (R,) is a cocycle and in this case, θ(x,y) is represented by e B ( x , y ) e M ( x y ) M ( x ) M ( y ) .

Let us consider the generalized characterization of a symmetrically compositive sum form related to distance measures with a cocycle:

(CDM) f(pr,qs)+f(ps,qr)=θ(pq,rs)f(p,q)f(r,s)

for all p,q,r,sG and where f, θ are functionals on G 2 , which can be represented as exponential functional equation in reduction.

In fact, if f(x,y)= 1 x + 1 y , then f(pr,qs)+f(ps,qr)=f(p,q)f(r,s), and also if f(x,y)= a ln x y , and θ(x,y)=2 then f, θ satisfy the equation f(pr,qs)+f(ps,qr)=θ(pq,rs)f(p,q)f(r,s).

This paper aims to investigate the superstability of four generalized functional equations of (CDM), namely, as well as that of the following type functional equations:

(G M f f f g ) f(pr,qs)+f(ps,qr)=θ(pq,rs)f(p,q)g(r,s),

(G M f f g f ) f(pr,qs)+f(ps,qr)=θ(pq,rs)g(p,q)f(r,s),

(G M f f g g ) f(pr,qs)+f(ps,qr)=θ(pq,rs)g(p,q)g(r,s),

(G M f f g h ) f(pr,qs)+f(ps,qr)=θ(pq,rs)g(p,q)h(r,s).

2 Superstability of the equations

In this section, we investigate the superstability of (CDM) and four generalized functional equations (G M f f f g ), (G M f f g f ), (G M f f g g ), and (G M f f g h ).

Theorem 1 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) h ( r , s ) | ϕ(r,s)p,q,r,sG.
(2.1)

and |f(p,q)g(p,q)|M for all p,qG and some constant M.

Then either g is bounded or h satisfies (CDM).

Proof Let g be an unbounded solution of inequality (2.1). Then there exists a sequence {( x n , y n )|nN} in G 2 such that 0|g( x n , y n )| as n.

Letting p= x n , q= y n in (2.1) and dividing by |θ( x n y n ,rs)g( x n , y n )|, we have

| f ( x n r , y n s ) + f ( x n s , y n r ) θ ( x n y n , r s ) g ( x n , y n ) h ( r , s ) | ϕ ( r , s ) k | g ( x n , y n ) | .

Passing to the limit as n, we obtain

h(r,s)= lim n f ( x n r , y n s ) + f ( x n s , y n r ) θ ( x n y n , r s ) g ( x n , y n ) .
(2.2)

Letting p= x n p, q= y n q in (2.1) and dividing by |g( x n , y n )|, we have

| f ( x n p r , y n q s ) + f ( x n p s , y n q r ) g ( x n , y n ) θ ( x n p y n q , r s ) g ( x n p , y n q ) g ( x n , y n ) h ( r , s ) | ϕ ( r , s ) | g ( x n , y n ) | 0
(2.3)

as n.

Letting p= x n q, q= y n p in (2.1) and dividing by |g( x n , y n )|, we have

| f ( x n q r , y n p s ) + f ( x n q s , y n p r ) g ( x n , y n ) θ ( x n q y n p , r s ) g ( x n q , y n p ) g ( x n , y n ) h ( r , s ) | ϕ ( r , s ) | g ( x n , y n ) | 0
(2.4)

as n.

Note that for any a, b, c in G, θ(ba,c)θ(b,a)=θ(b,ac)θ(a,c) by the definition of the cocycle. Letting pq=a, x n y n =b, and rs=c we have

θ ( x n y n p q , r s ) θ ( x n y n , p q ) θ ( x n y n , p q r s ) =θ(pq,rs)

for any p, q, r, s, x n , y n in G. Thus, from (2.2), (2.3), and (2.4), we obtain

| h ( p r , q s ) + h ( p s , q r ) θ ( p q , r s ) h ( p , q ) h ( r , s ) | = lim n | f ( x n p r , y n q s ) + f ( x n q s , y n p r ) + f ( x n p s , y n q r ) + f ( x n q r , y n p s ) θ ( x n y n , p r q s ) g ( x n , y n ) θ ( p q , r s ) h ( p , q ) h ( r , s ) | lim n | 1 θ ( x n y n , p r q s ) | | f ( x n p r , y n q s ) + f ( x n p s , y n q r ) g ( x n , y n ) θ ( x n p y n q , r s ) g ( x n p , y n q ) h ( r , s ) g ( x n , y n ) | + lim n | 1 θ ( x n y n , p r q s ) | | f ( x n q r , y n p s ) + f ( x n q s , y n p r ) g ( x n , y n ) θ ( x n q y n p , r s ) g ( x n q , y n p ) h ( r , s ) g ( x n , y n ) | + | h ( r , s ) | lim n | θ ( x n y n p q , r s ) θ ( x n y n , p q ) θ ( x n y n , p q r s ) g ( x n p , y n q ) + g ( x n q , y n p ) θ ( x n y n , p q ) g ( x n , y n ) θ ( p q , r s ) h ( p , q ) | h ( r , s ) θ ( p q , r s ) lim n | f ( x n p , y n q ) + f ( x n q , y n p ) θ ( x n y n , p q ) g ( x n , y n ) + ( g f ) ( x n p , y n q ) + ( g f ) ( x n q , y n p ) θ ( x n y n , p q ) g ( x n , y n ) h ( p , q ) | h ( r , s ) θ ( p q , r s ) lim n | 2 M k g ( x n , y n ) | + h ( r , s ) θ ( p q , r s ) lim n | f ( x n p , y n q ) + f ( x n q , y n p ) θ ( x n y n , p q ) g ( x n , y n ) h ( p , q ) | = 0 .

 □

Theorem 2 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) h ( r , s ) | ϕ(p,q)p,q,r,sG,
(2.5)

and |f(p,q)h(p,q)|M for all p,qG and some constant M.

Then either h is bounded or g satisfies (CDM).

Proof For h to be an unbounded solution of inequality (2.5), we can choose a sequence {( x n , y n )|nN} in G 2 such that 0|h( x n , y n )| as n.

Letting r= x n , s= y n in (2.5) and dividing by |θ(pq, x n y n )h( x n , y n )|, we have

| f ( p x n , q y n ) + f ( p y n , q x n ) θ ( p q , x n y n ) h ( x n , y n ) g ( p , q ) | ϕ ( p , q ) k | h ( x n , y n ) | .

Passing to the limit as n, we obtain

g(p,q)= lim n f ( p x n , q y n ) + f ( p y n , q x n ) θ ( p q , x n y n ) h ( x n , y n ) .
(2.6)

Replacing r=r x n , s=s y n in (2.5) and dividing by |h( x n , y n )|, we have

| f ( p r x n , q s y n ) + f ( p s y n , q r x n ) h ( x n , y n ) θ ( p q , r x n s y n ) g ( p , q ) h ( r x n , s y n ) h ( x n , y n ) | ϕ ( p , q ) | h ( x n , y n ) | 0
(2.7)

as n.

Replacing r=r y n , s=s x n in (2.5) and dividing by |h( x n , y n )|, we have

| f ( p r y n , q s x n ) + f ( p s x n , q r y n ) h ( x n , y n ) g ( p , q ) θ ( p q , r y n s x n ) h ( r y n , s x n ) h ( x n , y n ) | ϕ ( p , q ) | h ( x n , y n ) | 0
(2.8)

as n.

Thus from (2.6), (2.7), and (2.8), we obtain

| g ( p r , q s ) + g ( p s , q r ) θ ( p q , r s ) g ( p , q ) g ( r , s ) | = lim n | f ( p r x n , q s y n ) + f ( p r y n , q s x n ) + f ( p s x n , q r y n ) + f ( p s y n , q r x n ) θ ( p r q s , x n y n ) h ( x n , y n ) θ ( p q , r s ) g ( p , q ) g ( r , s ) | lim n | 1 θ ( p q r s , x n y n ) | | f ( p r x n , q s y n ) + f ( p s y n , q r x n ) h ( x n , y n ) g ( p , q ) θ ( p q , r x n s y n ) h ( r x n , s y n ) h ( x n , y n ) | + lim n | 1 θ ( p q r s , x n y n ) | | f ( p r y n , q s x n ) + f ( p s x n , q r y n ) h ( x n , y n ) g ( p , q ) θ ( p q , r y n s x n ) h ( r y n , s x n ) h ( x n , y n ) | + | g ( p , q ) | lim n | θ ( p q , r x n s y n ) θ ( r s , x n y n ) θ ( p q r s , x n y n ) h ( r x n , s y n ) + h ( r y n , s x n ) θ ( r s , x n y n ) h ( x n y n ) θ ( p q , r s ) g ( r , s ) | = | g ( p , q ) | θ ( p q , r s ) lim n | ( h f ) ( r x n , s y n ) + ( h f ) ( r y n , s x n ) θ ( r s , x n y n ) h ( x n , y n ) + f ( r x n , s y n ) + f ( r y n , s x n ) θ ( r s , x n y n ) h ( x n , y n ) g ( r , s ) | | g ( p , q ) | θ ( p q , r s ) 2 M k | h ( x n , y n ) | + | g ( p , q ) | θ ( p q , r s ) lim n | f ( r x n , s y n ) + f ( r y n , s x n ) θ ( r s , x n y n ) h ( x n , y n ) g ( r , s ) | = 0 .

 □

Corollary 1 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) g ( r , s ) | ϕ(p,q) or ϕ(r,s)

for any p,q,r,sG and |f(p,q)g(p,q)|M for all p,qG and some constant M. Then either g is bounded or g satisfies (CDM).

Corollary 2 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) f ( p , q ) g ( r , s ) | ϕ(p,q)

for any p,q,r,sG. Then either g is bounded, or f satisfies (CDM) and also f and g satisfy (G M f f f g ).

Corollary 3 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) f ( p , q ) g ( r , s ) | ϕ(r,s)

for any p,q,r,sG. Then either f is bounded, or g satisfies (CDM) and also g and f satisfy

(G M g g g f ) g(pr,qs)+g(ps,qr)θ(pq,rs)g(p,q)f(r,s).

Corollary 4 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) f ( r , s ) | ϕ(p,q)p,q,r,sG

for any p,q,r,sG. Then either f is bounded, or g satisfies (CDM) and also f and g satisfy (G M g g g f ).

Corollary 5 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) f ( r , s ) | ϕ(r,s)p,q,r,sG

for any p,q,r,sG. Then either g is bounded, or f satisfies (CDM) and also f and g satisfy (G M f f f g ).

Corollary 6 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) f ( r , s ) | ϕ(p,q)p,q,r,sG

for any p,q,r,sG. Then either f is bounded, or g satisfies (CDM) and also f and g satisfy (G M g g g f ).

Corollary 7 Let k>0 and f,g: G 2 R, ϕ: G 2 R + be functions satisfying

| f ( p r , q s ) + f ( p s , q r ) k ln ( p q ) ln ( r s ) f ( p , q ) f ( r , s ) | ϕ(p,q) or ϕ(r,s)

for any p,q,r,sG. Then either f is bounded or f satisfies the following equation:

f(pr,qs)+f(ps,qr)= k ln ( p q ) ln ( r s ) f(p,q)f(r,s).

Corollary 8 Let f,g: G 2 R, ϕ: G 2 R + be functions satisfying

| f ( p r , q s ) + f ( p s , q r ) f ( p , q ) f ( r , s ) | ϕ(p,q) or ϕ(r,s)

for any p,q,r,sG. Then either f is bounded or f satisfies (FE).

Theorem 3 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) f ( p , q ) g ( r , s ) | ε

for any p,q,r,sG. Then f (or g) is bounded, or f and g satisfy (CDM) and also f, g, θ satisfy (G M f f f g ).

Proof Replacing g(p,q) by f(p,q) and h(r,s) by g(r,s) for all p,q,r,sG in Theorem 1, we find that f is bounded or g satisfies (CDM). Note that f is bounded iff g is bounded. Namely, for all p,q,r,sG

| g ( r , s ) | ε + f ( p r , q s ) + f ( p s , q r ) k | f ( p , q ) | .

Let g be unbounded. Then f is unbounded by a similar method to the proof of Theorem 1; g satisfies (CDM). Now by a similar method to the calculation in Theorem 1 with the unboundedness of g, we have

f(p,q)= lim n f ( p x n , q y n ) + f ( p y n , q x n ) θ ( p q , x n y n ) g ( x n , y n )

for any r,s, x n , y n G. Since g satisfies (CDM), we have

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) f ( p , q ) g ( r , s ) | = lim n | f ( p r x n , q s y n ) + f ( p r y n , q s x n ) + f ( p s x n , q r y n ) + f ( p s y n , q r x n ) θ ( p r q s , x n y n ) g ( x n , y n ) θ ( p q , r s ) f ( p , q ) g ( r , s ) | lim n | 1 θ ( p q r s , x n y n ) | | f ( p r x n , q s y n ) + f ( p s y n , q r x n ) g ( x n , y n ) f ( p , q ) θ ( p q , r x n s y n ) g ( r x n , s y n ) g ( x n , y n ) | + lim n | 1 θ ( p q r s , x n y n ) | | f ( p r y n , q s x n ) + f ( p s x n , q r y n ) g ( x n , y n ) f ( p , q ) θ ( p q , r y n s x n ) g ( r y n , s x n ) g ( x n , y n ) | + | f ( p , q ) | lim n | θ ( p q , r x n s y n ) θ ( r s , x n y n ) θ ( p q r s , x n y n ) g ( r x n , s y n ) + g ( r y n , s x n ) θ ( r s , x n y n ) g ( x n y n ) θ ( p q , r s ) g ( r , s ) | = | f ( p , q ) | lim n | θ ( p q , r x n s y n ) θ ( r s , x n y n ) θ ( p q r s , x n y n ) g ( r x n , s y n ) + g ( r y n , s x n ) θ ( r s , x n y n ) g ( x n y n ) θ ( p q , r s ) g ( r , s ) | = | f ( p , q ) | | θ ( p q , r s ) g ( r , s ) θ ( p q , r s ) g ( r , s ) | = 0 .

Thus f and g imply the required (G M f f f g ). The same procedure implies that the above inequalities change to

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) f ( p , q ) f ( r , s ) | | f ( p , q ) | lim n | θ ( p q , r x n s y n ) θ ( r s , x n y n ) θ ( p q r s , x n y n ) f ( r x n , s y n ) + f ( r y n , s x n ) θ ( r s , x n y n ) g ( x n y n ) θ ( p q , r s ) f ( r , s ) | = | f ( p , q ) | | θ ( p q , r s ) f ( r , s ) θ ( p q , r s ) f ( r , s ) | = 0 ,

as desired. □

The proof of the following theorem is the same procedure as in the proof of Theorem 3.

Theorem 4 Let f,g: G 2 R, ϕ: G 2 R + be functions and a function θ: G 2 R k be a cocycle satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) f ( r , s ) | ε

for any p,q,r,sG. Then f (or g) is bounded, or f and g satisfy (CDM) and also f, g, θ satisfy (G M f f f g ).

Example 1 Let

f(x,y)= a ln x y + ε 2 ,g(x,y)= a ln x y ,θ(x,y)=2.

Then we have

| f ( p , q ) g ( p , q ) | ε 2

and

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) g ( p , q ) g ( r , s ) | = | a ln p r q s + a ln p s q r + ε 2 a ln p q a ln r s | = ε .

Thus g satisfies (CDM). But f, g, θ being nonzero functions do not satisfy (G M f f g g ).

Let (S;) and ( S ˜ ;) be a semigroup and a group with semigroup operation ⋄, respectively.

Theorem 5 Let f,g,h: S 2 , S ˜ 2 R and ϕ: S 2 , S ˜ 2 R be a nonzero function satisfying

| f ( p r , q s ) + f ( p s , q r ) θ ( p q , r s ) f ( p , q ) g ( r , s ) | { i ( i ) ϕ ( r , s ) p , q , r , s S ˜ , ( ii ) ϕ ( p , q ) p , q , r , s S .
(2.9)
  1. (a)

    In case (i), let |f(p,q)g(p,q)|M for all p,qS and some constant M.

    Then either g is bounded or h satisfies (CDM).

  2. (b)

    In case (ii), let |f(p,q)h(p,q)|M for all p,qG and some constant M.

Then either h is bounded or g satisfies (CDM).