Quote:
Originally Posted by Ratamahatta
Re: the first integral:
I don't like your argument about a limit; it's not convincing. You should be more precise, though your conclusion is correct. An inequality would be easiest (and it is quite easy). In the divergent case, you should bound the integral below by something known to diverge. In the convergent case, you should bound the integral above by something known to converge.
Re: the second integral:
For simplicity, let p = -a/sigma^2, and let q = b/sigma^2. Assume q > 0.
Exponentials dominate polynomials, so for some large enough value X, we have the following:
x^p < e^(qx/2) for all x > X.
Now, then our integral converges iff the integral from X to infinity converges, but the integral from X to infinity is <= the integral of e^(qx/2)*e^(-qx) = Integral of e^(-qx/2), which converges.
So this converges when b > 0.
If b is negative (and thus e is being raised to a positive power),
then if p > 0, then for large enough x, both x^p and e^(-qx) are greater than 1. Since the integrand does not go to 0, the integral cannot converge.
If p < 0, then notice that for some large enough value X, we have:
e^(-qx) > x^(-p) for all x > X.
So again, the limit of the integrand (as x --> infinity) is > 0, so the integral cannot converge.
hth