Expectation of dependent Bernoulli sum

219 Views Asked by At

I want to estimate the expected value of the following sum of random variables,

$$ Y = \sum_{i=1}^N X_i $$

where each $X_i$ is a Bernoulli random variable. In particular,

$$ X_1 = \begin{cases} 1, & \text{with prob. } 1/N \\ 0, & \text{with prob. } 1 - 1/N \end{cases} $$

and for all $2 \leq i \leq N$

$$ X_i = \begin{cases} 1, & \text{with prob. } p_i \\ 0, & \text{with prob. } 1 - p_i \end{cases} $$

where

$$ p_i = \begin{cases} p_{i-1} + 1/N, & \text{if } X_{i-1} = 0 \\ 1/N, & \text{if } X_{i-1} = 1 \end{cases} $$

I have computed the value for $ N=2 $ and $ N=3 $. For $ N = 2 $,

$$ \mathbb{E}[Y] = 1 \left[ \left( 1 - \frac{1}{N} \right) \frac{2}{N} + \frac{1}{N} \left( 1 - \frac{1}{N} \right) \right] + 2 \frac{1}{N^2} = \frac{5}{4} $$

where the first term corresponds to the sequences $(01)$ and $(10)$, while the second to $(11)$.

For $ N=3 $, here are the calculations

$$ \begin{gather} \mathbb{E}[Y] = \left(1-\frac{1}{N}\right)\left(1-\frac{2}{N}\right)\frac{3}{N} + 2 \times \left(1-\frac{1}{N}\right)\frac{2}{N} \frac{1}{N} + 2 \times \frac{1}{N}\left(1-\frac{1}{N}\right)\frac{2}{N} + 3 \times \frac{1}{N} \frac{1}{N} \frac{1}{N} + \left(1 - \frac{1}{N}\right)\frac{2}{N}\left(1-\frac{1}{N}\right) + \frac{1}{N}\left(1-\frac{1}{N}\right)\left(1-\frac{2}{N}\right) + 2 \times \frac{1}{N} \frac{1}{N} \left(1-\frac{1}{N}\right) \\ = \frac{3}{N^3} - \frac{6}{N^2} + \frac{6}{N} \\ N=3 \rightarrow = \frac{39}{27} \end{gather} $$

This results from the sequences (001), (011), (101), (111), (010), (100), (110).

Is there any way to proceed farther?

2

There are 2 best solutions below

1
On

MY TRY

We first look at $ P\{ Y=i\}$, i.e., $Y$ has exactly $i$ ones and $N-i$ zeros. The probability corresponding to ones is $N^{-i}$. The probability corresponding to zeros depends on number and size of blocks of zeros. There can be at most $i+1$ blocks of size between 0 and $N-i$, as we already have $i$ ones. Further, the probability corresponding to a block of size $a_k$ is $(a_k+1)!N^{-a_k} $. Thus, we have \begin{align} P\{ Y=i\} &= N^{-i}\prod_{\substack{a_1,a_2,\ldots,a_{i+1}\geq 0\\ \sum_{k=1}^{i+1} a_k=N-i} }1-\frac{(a_k+1)!}{N^{a_k}}\\ &= N^{-N}\prod_{\substack{a_1,a_2,\ldots,a_{i+1i}\geq 0\\ \sum_{k=1}^{i+1} a_k=N-i} } N^N-(a_k+1)! \end{align} Also, we have the following: \begin{align} E\{Y\} = \sum_{i=0}^N i P\{ Y=i\} = N^{-N} \sum_{i=0}^N i \prod_{\substack{a_1,a_2,\ldots,a_{i+1}\geq 0\\ \sum_{k=1}^{i+1} a_k=N-i} } N^N-(a_k+1)! \end{align} I don't know how to simplify it further!

0
On

Not an answer / too long for a comment.

Here are some numerical results:

  N    E[Y]
  1  1.000000
  2  1.250000
  3  1.444444
  4  1.621094
  5  1.785920
  6  1.940608
  7  2.086209
  8  2.223695
  9  2.353993
 10  2.477955
 11  2.596337
 12  2.709795
 13  2.818884
 14  2.924076
 15  3.025769
 16  3.124298
 17  3.219949
 18  3.312966
 19  3.403559
 20  3.491909
 21  3.578177
 22  3.662504
 23  3.745014
 24  3.825820
 25  3.905022
 26  3.982712
 27  4.058972
 28  4.133878

The growth rate seems to be $E[Y] \propto N^{0.4494}$, so an upperbound of $\sqrt{N}$ seems plausible.