Scribd
Upload
26 views

Memory Less

Math

Uploaded by

NguyễnHuyHùng
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
26 views

Memory Less

Math

Uploaded by

NguyễnHuyHùng
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Notes on Memoryless Random Variables

Math 468 / 568 Spring 2010


[email protected]

A random variable X is memoryless if for all numbers a and b in its range, we have
P (X > a + b|X > b) = P (X > a) .

(1)

(We are implicitly assuming that whenever a and b are both in the range of X, then so is a + b. The
memoryless property doesnt make much sense without that assumption.) It is easy to prove that if
the range of X is [0, ), then X must be exponential. Similarly, if the range of X is {0, 1, 2, ...},
then X must be geometric.
Proof for the [0, ) case: Lets define G(a) := P (X > a). If X is memoryless, then G has the
following properties:
1. G(a + b) = G(a) G(b).
2. G is monotonically decreasing, i.e., if a b then G(a) G(b).
Property 1 follows directly from Eq. (1); Property 2 is a consequence of the definition of G. Well
see that the above two conditions imply that for all real x 0,
G(x) = G(1)x .

(2)

Setting = log(G(1)) would then gives G(x) = P (X > x) = ex , i.e., X is exponential.


To prove Eq. (1), first assume x is rational, i.e., x = m/n for integers m and n. Using
Property 1 above, we see that G(m/n) = G(1/n)m . Raising both sides to the nth power yields
G(m/n)n = (G(1/n)m )n = (G(1/n)n )m = G(1)m , where the last step uses Property 1 again.
Taking the 1/nth power of both sides gives G(m/n) = G(1)m/n .
Now let x be any real number 0. Choose any two sequences qn and rn of rational numbers
such that qn < x < rn for all n and qn , rn x as n . From Property 2 above, we get
G(1)qn = G(qn ) G(x) G(rn ) = G(1)rn .
Letting n gives G(1)x = G(x), as desired.

(3)
QED

Property 1 above is equivalent to Cauchys functional equation:


g(a + b) = g(a) + g(b) ;

(4)

this is obtained from Property 1 by setting g(x) = log(G(x)). The proof above shows that if g is
decreasing, then the only solution of Eq. (4) is g(x) = cx for some constant c. What happens if we
drop the requirement that g is decreasing? Then there can be extremely strange solutions to Eq. (4)!
These pathological solutions have some amazing properties, including 1) discontinuities at every
point; 2) unbounded on every open interval; and 3) they are not measurable. Such functions
cannot be the pdf of any random variable!
If youre interested in learning more, start with
http://en.wikipedia.org/wiki/Cauchy functional equation
You might also want to check out Lectures on Functional Equations and Their Applications by J
Aczel.

You might also like