Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I'm looking for a convenient upper bound on the integral $\int_y^\infty x^k \exp(-(x-\mu)^2/2) dx$ for (possibly large) positive integer k. This is equivalent to finding higher moments of a truncated normal distribution. A bound that works for non-integer k as well would be even better.

Of course "convenient" is in the eye of the beholder, but I'd like some sort of fairly simple expression that I can use in further calculations. For example, an upper bound of the form $f(x) \exp( -g(x))$ where where $f$ and $g$ are low-degree polynomials would be great. I'm more interested in simplicity of form than in obtaining the tightest possible bound.

share|improve this question

2 Answers 2

up vote 3 down vote accepted

I assume $y \gt 0$ and $y \gg \mu$. If you replace $x^k$ by $y^k \exp {\left( (k/y)x - k \right) }$ you will overestimate the $x^k$ term (because this is the exponential of the first two terms of the MacLaurin series of $\log(x^k)$ expanded around $x=y$, the series is alternating, and the remainder term is negative). Completing the square yields a closed-form formula for an upper bound, one of whose factors is a Gaussian integral:

$$\sqrt{2\pi }y^k \exp \left( {\frac{k (k+2 y (-y+\mu ))}{2 y^2}} \right) \Phi \left(\frac{k}{y}-y+\mu \right)$$

This will work extremely well when $k$ is large compared to $y$ and $\mu$ because then most of the mass of the integral is concentrated at its lower limit where the exponential upper bound to $x^k$ is a good approximation. To avoid exponential overflow, use logarithms to compute the product.

share|improve this answer
    
Thanks! That's clever and useful. –  John D. Cook Oct 4 '10 at 11:22

Let $I_k=\int_y^\infty x^k\mathrm{exp}(-(x-\mu)^2/2)dx$ and $a_k=y^{k}\mathrm{exp}(-(y-\mu)^2/2)$.

Integration by parts gives $I_k=a_{k-1}+(k-1)I_{k-2}$.

Therefore, for even $n$, \begin{multline} I_n=a_{n-1}+(n-1)a_{n-3}+(n-1)(n-3)a_{n-5}+\cdots\newline+(n-1)(n-3)\cdots 1 a_1+(n-1)(n-3)\cdots 1 I_0. \end{multline}

Feel free to expand this answer and deduce an estimate from the above.

share|improve this answer
    
Thanks. I've updated the question to be more specific regarding what I'm looking for. –  John D. Cook Sep 30 '10 at 14:50
    
What happened to $\mu$? $I_n$ clearly must include a polynomial in $\mu$ times the exponential, so the recursion coefficients have to depend on $\mu$ as well as on $n$. –  whuber Oct 1 '10 at 18:18
    
@whuber: You are right. What I wrote only holds for $\mu=0$. –  Rasmus Oct 2 '10 at 15:43

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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