Baruah (2011b) has successfully shown the construction of a normal imprecise number with the help of the operation of superimposition of real intervals. Das et al. (2013) has shown the construction of normal imprecise number using data from earthquake waveform and has studied the pattern of the membership curve of the waveform. In this article, instead of superimposing real intervals, it will be discussed about how to obtain the membership surface of a two dimensional imprecise vector if we superimpose some plates in the two dimensional plane. In Figure 1 two plates are superimposed restricting the condition that the intersection of the two plates is not void.
We can easily visualize in Figure 1, that the probability of the shaded area is 1 and the probability for the unshaded area of the plates will be ½. But if the number of superimposition is large then it will be very difficult to obtain the probabilities by simply observing the imposition of the plates. So, in that situation a different technique can be used to obtain the probabilities when the number of operation of superimposition is very large. At first, it is discussed, about the operation of superimposition in the two dimensional case when the variable X is imprecise but Y is not imprecise.
The operation of superimposition
The operation of superimposition of two real intervals [(a1, 0), (b1, 0)] and [(a2, 0), (b2, 0)] as
where (a(1), 0) = min[(a1, 0), (a2, 0)] , (a(2), 0) = max[(a1, 0), (a2, 0)] , (b(1), 0) = min[(b1, 0), (b2, 0)] and (b(2), 0) = max[(b1, 0), (b2, 0)]. Here we have assumed without any loss of generality that [(a1, 0), (b1, 0)] ∩ [(a2, 0), (b2, 0)] is not void or in other words that max[(a
i
, 0)] ≤ min[(b
i
, 0)], i = 1, 2.
For the three intervals and all with elements with a constant level of partial presence equal to 1/3 everywhere (See Figures 2, 3 and 4), we shall have
where, for example represents the interval [(y(1), 0), (y(2), 0)] with level of partial presence 2/3 for all elements in the entire interval, (x(1), 0), (x(2), 0), (x(3), 0) be the values of (x1, 0), (x2, 0), (x3, 0) arranged in increasing order of magnitude, and similarly (y(1), 0), (y(2), 0), (y(3), 0) be the values of (y1, 0), (y2, 0), (y3, 0) arranged in increasing order of magnitude again. We here presumed that [(x1, 0), (y1, 0)] ∩ [(x2, 0), (y2, 0)] ∩ [(x3, 0), (y3, 0)] is not void. It can be seen that for n imprecise intervals , all with membership value are equal to everywhere, we shall have
where for example represents the uniformly imprecise interval [(b(1), 0), (b(2), 0)] with membership in the entire interval, (a(1), 0), (a(2), 0), …, (a(n), 0) be the values of (a1, 0), (a2, 0), …, (a
n
, 0) arranged in increasing order of magnitude and (b(1), 0), (b(2), 0), …, (b(n), 0) be the values of (b1, 0), (b2, 0), …, (b
n
, 0) arranged in increasing order of magnitude. Thus for the imprecise intervals all with uniform membership , the values of membership of the superimposed imprecise intervals are and . These values of membership considered in two halves as
and
would suggest that they can define an empirical distribution and a complementary empirical distribution on (x11, 0), (x12, 0), …, (x1n, 0) and (x21, 0), (x22, 0), …, (x2n, 0) respectively. In other words, for realizations of the values of (x(11), 0), (x(12), 0), …, (x(1n), 0) are in increasing order and of (x(21), 0), (x(22), 0), …, (x(2n), 0) again are in increasing order, we can see that if we define
then the Glivenko – Cantelli Lemma on Order Statistics assures that
where ∏1[(a, 0), (x1, 0)], (a, 0) ≤ (x1, 0) ≤ (b, 0) and ψ2(x
2
, 0), (b, 0) ≤ (x
2
, 0) ≤ (c, 0) ψ2(x2, 0), (b, 0) ≤ (x2, 0) ≤ (c, 0) are two probability distributions. Thus
where
Thus, if φ1(x, 0) and (1 - φ2(x, 0)) are two independent probability distribution functions defined in [(α, 0), (β, 0)] and [(β, 0), (γ, 0)] respectively, then the membership surface of a normal imprecise vector N = [(α, 0), (β, 0), (γ, 0)] can be expressed as
or
Here, in the case of two dimensions we have considered that the value of y is zero. But instead of zero if we consider any precise value of y, then in the above membership surface only the value of y will be changed.
Similarly, we can also show that, the membership surface of a normal imprecise vector N = [(0, α), (0, β), (0, γ)] can be expressed as
Here, we have also considered the value of x is zero. But instead of zero if we consider any precise value of x, then in the above membership surface only the value of x will be changed.
Now, we are going to discuss about the method how to obtain the membership surface of the vector (X, Y), where the variables x and y both are imprecise.
Consider an imprecise vector (X, Y), where X and Y are imprecise represented by X = [a, b, c] and Y = [p, q, r] respectively. Assume that X and Y are independently distributed. Let the membership function of X and Y be μX|Y(x, y) and μY|X(x, y) as mentioned below
and
Then the membership surface of the imprecise vector (X, Y) can be obtained as follows
For a three dimensional imprecise vector (X, Y, Z), where the membership functions of X, Y and Z are as mentioned below:
Then the membership surface of the imprecise vector (X, Y, Z) will be as shown below: