2
$\begingroup$

I am trying to find following expectation

\begin{equation} \mathbb{E} [\frac{1}{1+X}] \end{equation}

where $X$ is a non central Chi Squared distribution with $K$ degree of freedom and $K$ non-centrality parameter. I know it must be related to Marqum Q function however I can't find something like this in integral tables and I don't have Mathematica.

$\endgroup$
  • $\begingroup$ @wolfies how did you find that? actually I needed a closed form solution to address my problem. originally I needed the solution for $\mathbb{E}[\frac{1}{1+a X}]$ , where $a$ is a positive real number $\endgroup$ – Alireza Aug 14 '17 at 16:34
  • $\begingroup$ @wolfies hello? $\endgroup$ – Alireza Aug 18 '17 at 21:56
  • $\begingroup$ I've posted some thoughts below $\endgroup$ – wolfies Aug 19 '17 at 17:39
1
$\begingroup$

To define notation, let $X \sim \text{NoncentralChisquared}(k,k)$ with pdf $f(x)$:

enter image description here

Symbolic solution to $\mathbb{E} [\frac{1}{1+X}]$

A closed form solution to $\mathbb{E} [\frac{1}{1+X}]$ does not seem readily obtainable, whether by transforming the pdf to $Y = 1/(1+X)$ or otherwise.

The neatest form I have obtained is first by deriving the mgf $M_X(t) = \mathbb{E}[e^{t X}]$ as:

enter image description here

where I am using the Expect function from mathStatica/Mathematica to automate the nitty gritties (I should note I am one of the authors of the function).

The mgf can be used to derive this negative moment (for non-negative random variables) via:

$$\mathbb{E} [\frac{1}{1+ aX}] \quad = \quad \int_0^{\infty } e^{-t} M(-a t) \;dt$$

When $a = 1$, this is: $$\mathbb{E}[\frac{1}{1+X}] \quad = \quad \int_0^{\infty } \exp\big(-\frac{k t}{1+2t}-t\big) (1+2t)^{-k/2} \;dt$$

which can be solved for odd-valued integers.

For example, when $k = 1$, the exact solution is: $$\frac{e^{-i} \sqrt{\pi } \left(\text{erf}\left((-1)^{3/4}\right)+e^{2 i} \text{erfc}\left(\sqrt[4]{-1}\right)+1\right)}{2 \sqrt{2}} \quad \approx \quad 0.520862$$

Approximation

The above seems far too messy. Fortunately, an exceptional and simple approximation is: $$\mathbb{E} [\frac{1}{1+X}] \; \; \approx \; \frac{1}{2(k-1)} \quad \quad \text{for} \quad k> 7$$

The following figure compares:

  • $\mathbb{E} [\frac{1}{1+X}]$ (calculated by numerical integration) - BLUE curve
  • ... to the simple $\frac{1}{2(k-1)}$ approximation - ORANGE dashed curve

enter image description here

For $k> 10$, the approximation appears to be accurate to at least 3 decimal places.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.