Information Technology Reference
In-Depth Information
if
X
and
Y
are independent random variables, from the definitions of entropy, joint
entropy, conditional entropy, and random variable independence, it follows that the
entropy of
Z
is the sum of the entropies of its components
X
and
Y
.
According to Shannon's information theory, the entropy of probability dis-
tributions with a given variance reaches the maximum value in the normal
distribution having that variance (see Table 4.8 for a proof in the continuous
case). Therefore, the joint probability distribution of
Z
=
√
X
2
Y
2
reaches its
maximum, when both components
X
and
Y
reach their maxima. As we have
shown above, the Pythagorean recombination game transforms a distribution
in another one having the same variance, therefore if the distributions of the
two Pythagorean components of these distributions evolve toward normal dis-
tributions, then, along the game, the
H
function evolves toward its minimum
value (and
S
toward its maximum value).
+
1
2
ln
2
The entropy of a normal distribution is
(
2
π
e
σ
)
, as deduced in Table 4.7.
2
Ta b l e 4 . 7
The entropy of the Gaussian distribution
N
of mean 0 and variance
σ
+
∞
S
(
N
)=
−
N
(
x
)
ln
N
(
x
)
dx
−
∞
x
2
2σ
+
∞
−
∞
−
N
(
x
)
ln
e
−
2
√
2πσ
=
2
dx
+
∞
−
∞
−
ln
2
x
2
2σ
=
N
(
x
)[
−
2
−
πσ
2
]
dx
+
∞
2
+
∞
−
∞
dx
+
ln
2πσ
N
(
x
)
x
2
2σ
=
N
(
x
)
dx
2
−
∞
2
+
ln
2πσ
E
(
x
2
)
2
=
·
1
2σ
1
1
2
=
2
+
2
ln 2πσ
1
2
(
1
+
ln 2πσ
2
=
)
1
2
(
ln
e
+
ln
(
2πσ
2
=
))
2
(
2
=
(
)
ln
2
π
e
σ
The
entropic divergence
(or
Kullback-Leibler divergence
) between two proba-
bility distributions
p
,
q
, denoted by
D
(
p
q
)
is given by: