Hello,

I need to compute the intersection probability of two intervals randomly
chosen in the domain [0,1] such to respect some constraints related to
their sizes. Precisely, the size of the first interval has to be between
a and b, and the size of the second interval between c and d,
where 0 <= a <= b <= 1 and 0 <= c <= d <= 1.

I've managed to compute this probability for the simple case where no
size restriction is imposed to any of the two intervals. It gives 2/3 =
0.666... which verifies experimentally.

I've also managed to compute the intersection probability when the size
of the first interval is restricted between a and b, while the size of
the second interval has no restriction. Obviously the corresponding
formula depends on a and b:

P = (2*a^3 + 2*b^3 - 4*a^2 - 4*b^2 - 4*a*b + 3*a + 3*b + 2) / (2 -
a - b) / 3

I've experimentally verified this formula and it works well.

Nevertheless, my computing method was mostly intuitive based on the
discretization of the domain [0,1] in N regions and on the
representation of the intervals as ordered pairs of numbers between 1
and N. Considering such interval pairs, I counted and expressed as
functions of N: (1) the number of combinations of interval pairs
overlapping each other; and (2) the total number of possible interval
pair combinations. Considering N -> infinit, the fraction between the
two terms (1) / (2) gave me the intersection probability .

Unfortunately, my counting method becomes too complex when also
restricting the size of the second interval to [c,d].

I know that such problems could be better expressed as integrals in a
suitable multi-dimesional data space (probably 4-dimensional in this
case), but this is not trivial to me.

Could someone help me to express the interval intersection probability
in a mathematical form, such to compute the general formula depending on
a, b, c and d?

Thank you very much,

Cristian Saita

.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to