International Journal of Pure and Applied Mathematics

Volume 32 No. 4 2006, 467-473

A GENERALIZED USEFUL INACCURACY OF ORDER

AND TYPE AND CODING THEOREMS

M.A.K. Baig1 , Rayees Ahmad Dar2

1,2 Department of Statistics
University of Kashmir
Srinagar, 190006, INDIA
1 e-mail: [email protected]
2 e-mail: rayees [email protected]

Abstract: Useful inaccuracy measures and mean codeword lengths are well
known in the literature of information theorey. In this communication, a new
generalized useful inaccuracy of order and type has been proposed and
coding theorem has been established by considering the said measure and a
generalized average useful codeword length. Our motivation for studying this
is that it generalizes some results which already existing in the literature.

AMS Subject Classification: 94A17, 94A24

Key Words: inaccuracy measures, mean codeword length, Holders inequality

1. Introduction

Consider the model given below for a finite random experiment scheme hav-
ing (x1 , x2 , ..., xn ) as a complete system of events, happening with respective
probabilities P = (p1 , p2 , ..., pn ) and credited with utilities U = (u1 , u2 , ..., un ),
ui > 0, i = 1, 2, ..., n. Denote
 
= x1 x2 .......xn p1 p2 ......pn u1 u2.........un , (1.1)
we call (1.1) as utility information scheme.
Let Q = (q1 , q2 , ..., qn ) be the predicted distribution having the utility distri-
bution (u1 , u2 , ..., un ). Taneja and Tuteja [14] have suggested and characterized
the useful inaccuracy measure
Received: September 19, 2006 c 2006, Academic Publications Ltd.

Correspondence author
468 M.A.K. Baig, R.A. Dar

n
X
I (P ; Q; U ) = ui pi log qi . (1.2)
i=1

By considering weighted mean codeword length, see [6]

n
P
ui pi li
i=1
L (U ) = n . (1.3)
P
ui pi
i=1

Taneja and Tuteja [14] derived the lower and upper bounds on L (U ) interms
of I (P ; Q; U ).
Bhatia [3] defined the useful average code lengths of order as
1

n
1 ui
Dli ( 1

) ,
X
L (U ) = 1 pi (1.4)

1 n
D 1

P
ui pi

i=1
i=1

n
P
where > 0 (6= 1) and pi 1, i = 1, 2, ..., n and D is the size of the code
i=1
alphabet. He also derived the bounds for the useful average code length of
order and is given by
P n 1
1
upq
1 i=1 i i i
I (P ; Q; U ) = 1 1 , (1.5)

n
D 1
P
ui pi

i=1

n
P
where > 0 (6= 1) and pi 0, pi 1, i = 1, 2, ..., n and D is the size of the
i=1
code alphabet.
Under the condition
n
X
pi qi1 D li 1 , (1.6)
i=1

where D is the size of the code alphabet. Inequality (1.6) is a generalized

Krafts inequality [4]. A code satisfying generalized Krafts inequality would be
termed as personal probability code.
Longo [10], Gurdial and Pessoa [5], Autar and Khan [1], Jain and Tuteja
[8], Taneja et al [15], Hooda and Bhaker [7], Bhatia [3] and Singh, Kumar and
A GENERALIZED USEFUL INACCURACY OF ORDER... 469

Tuteja [11] considerd the problem of useful information measure and used it
studying the noiseless coding theorems for sources involving utilities.
In the next section, we shall study some coding theorems for a generalized
useful inaccuracy of order and type for incomplete probability distribution.

2. Coding Theorems

Consider the function

n
P 1

ui pi qi1
1 i=1
I (P ; Q; U ) = 1 , (2.1)

1 n
D 1

P
ui p i

i=1
n
P
where > 0 (6= 1), > 0, pi 0 and pi 1, i = 1, 2, ..., n and D is the size
i=1
of the code alphabet.
n
P
(a) For 1, = 1 and pi = 1, the measure (2.1) reduces to Taneja
i=1
and Tuteja [15] measure of useful inaccuracy.
(b) For = 1, pi = qi , i = 1, 2, ..., n, the measure (2.1) reduces to the
measure given by Autar and Khan [1] useful information measure.
(c) For 1, = 1 and pi = qi , i = 1, 2, ..., n, the measure (2.1)
reduces to Belis and Guiasu [2] measure of useful information for incomplete
probability distribution. Further, when utility aspect of the scheme is ignored,
the measure reduces to Shannon [12] measure of entropy.
(d) When the probability distribution is complete and the utility aspect of
the scheme is ignored as well as 1, = 1. The measure (2.1) becomes Ker-
ridges [9] measure of inaccuracy. We call (2.1) as generalized useful inaccuracy
of order and type for incomplete probability distribution.
Further, consider a generalized useful mean length credited with utilities
and probabilites as
1

n
1 X
ui li ( 1 )

L (U ) = 1 1 pi n D , (2.2)


D 1 ui pi
P
i=1
i=1
n
P
where > 0 (6= 1), > 0, pi 0 and pi 1, i = 1, 2, ..., n and D is the size
i=1
of the code alphabet.
470 M.A.K. Baig, R.A. Dar

(a) For 1, = 1, the measure (2.2) reduces to useful mean length

L (U ) of the code, given by Guiasu and Picard [6].
(b) when the utility aspect of the scheme is ignored by taking ui = 1,
n
P
i = 1, 2, .., n, pi = 1 and 1, = 1, the mean length (2.2) becomes
i=1
optimal code length identical to Shannon, see [12].
Now we find the bounds for L (U ) in terms of I (P ; Q; U ) under the
condition
Xn
pi qi1 D li 1 , (2.3)
i=1

where D is the size of the code alphabet.

Theorem 2.1. For all integers D (D > 1), let li satisfy (2.3), then the
generalized average useful codeword length satisfies

L (U ) I (P ; Q; U ) , (2.4)

and the equality holds iff

ui qi
li = log n
. (2.5)
ui pi qi 1
P
i=1

Proof. By Holders inequality, see [13]

n n
!1 n
!1
p q

xpi yiq
X X X
xi yi , (2.6)
i=1 i=1 i=1

where 1p + 1q = 1, p < 1 (6= 0) and xi , yi > 0, i = 1, 2, .., n, we see the equality

holds if and only if there exists a positive constant c such that

xpi = cyiq . (2.7)

Making the substitution

1
1

1 ui
p= , q = 1 , xi = pi1 n
Dli ,
P

ui pi
i=1
A GENERALIZED USEFUL INACCURACY OF ORDER... 471

1
1

ui
yi = pi1 n
P
qi1
ui pi
i=1

in (2.6), using
  (2.3) and after making suitable operations we get (2.4) for
1
D 1 6= 0 according as 6= 1.
Theorem 2.2. For every code with lengths {li } , i = 1, 2, ..., n of Theorem
2.1, L (U ) can satisfy the inequality
1
1 1D
L (U ) < I (P ; Q; U ) D + 1 . (2.8)
D 1

Proof. Let li be the positive integer satisfying the inequality

ui qi ui qi
log n
li < log
P 1 + 1 .
n
(2.9)
1
P
ui pi qi ui p i q i
i=1 i=1

Consider the intervals

ui qi ui qi
i = log
n
, log
n
+ 1 (2.10)
ui pi qi1 ui pi qi1
P P
i=1 i=1

of length 1. In every i , there lies exactly one positive integer li such that

ui qi
ui qi
0 < log n
li < log
n
+ 1. (2.11)
1 1
P P
ui p i q i ui pi qi
i=1 i=1

We will first show that the sequences l1 , l2 , ..., ln , thus defined satisfy (2.3).
From (2.11), we have

ui qi
log n
li ,
ui pi qi 1
P
i=1
472 M.A.K. Baig, R.A. Dar

or
ui qi li
P 1 D .

n
ui pi qi
i=1

Multiply both sides by pi qi1

and summing over i = 1, 2, .., n, we get (2.3).
The last inequality of (2.11) gives

ui qi
li < log n
+ 1,
ui pi qi 1
P
i=1
or 1
ui qi
Dli <

n
D,
1
P
ui p i q i
i=1
1

li 1
( )< ui qi 1
D
n
D .
1
P
ui p i q i
i=1
1

Multipying both sides by pi P

n
ui and summing over i, i = 1, 2, .., n
ui p
i
i=1
and after suitable operations, we get
1
1 1D
L (U ) < I (P ; Q; U ) D + 1 .  (2.12)
D 1
1
1D
Remark. For 0 < < 1 and since D 2, from (2.12), we have 1 >1
D 1
from which it follows that the upper bound of L (U ) in (2.8) is greater than
unity.

References

[1] R. Autar, A.B. Khan, On generalized useful information for incomplete

distribution, Journal of Combinatorics, Information and System Sciences,
14, No. 4 (1989), 187-191.
A GENERALIZED USEFUL INACCURACY OF ORDER... 473

[2] M. Belis, S. Guiasu, A qualitiative -quantitative measure of information in

cybernetics system, IEEE Transaction Information Theory, IT-14 (1968),
593-594.

[3] P.K. Bhatia, On a generalized useful inaccuracy for incomplete probability

distribution, Soochow Journal of Mathematics, 25, No. 2 (1999), 131-135.

[4] A. Feinstein, Foundation of Information Theory, McGraw Hill, New York

(1956).

[5] Gurdial, F. Pessoa, On useful information of order , Journal of Combi-

natorics, Information and System Sciences, 2 (1977).

[6] S. Guiasu, C.F. Picard, Borne inferieure dela langueur de certian codes,
C.R. Academic Sciences, Paris, 27C (1971), 248-251.

[7] D.S. Hooda, U.S. Bhaker, A generalized useful information measure and
coding theorem, Soochow Journal of Mathematics, 23, No. 1 (1997), 53-62.

[8] P. Jain, R.K. Tuteja, On coding theorem connected with useful entropy of
order , International Journal of Mathematics and Mathematical Sciences,
12, No. 1 (1989), 193-198.

[9] D.F. Kerridge, Inaccuracy and inference, Journal of Royal Statistical So-
ciety, 23B (1961), 184-194.

[10] G. Longo, Quantitative-Qualititative Measure of Information, Springer

Verlag, New York (1972).

[11] R.P. Singh, R. Kumar, R.K. Tuteja, Applications of Holders inequality in

information theory, Information Sciences, 152 (2003), 145-154.

[12] C.E. Shannon, A mathematical theory of communication, Bell System

Tech. Journal, 27, 394-423 (1948), 623-656.

[13] O. Shisha, Inequalities, Academic Press, New York (1967).

[14] H.C. Taneja, R.K. Tuteja, Characterization of a quantitative-qualitative

measure of inaccuracy, Kybernetika, 22 (1986), 393-402.

[15] H.C. Taneja, D.S. Hooda, R.K. Tuteja, Coding theorems on a generalized
useful information, Soochow Journal of Mathematics, 11 (1985), 123-131.
474