2
$\begingroup$

Moved this here from MathSE because I had no luck there and suspect the question may be harder than I first thought.

Consider the integral $$ I(\alpha)=\int_0^1 dx_1 \int_0^1 dy_1\int_{x_1}^1dx_2\int_{y_1}^1dy_2\,e^{-\alpha(x_2-x_1)(y_2-y_1)} $$ in the limit $\alpha\rightarrow\infty$. To find the asymptotic expansion in this case is not hard: you can evaluate the integral in closed form in terms of Exponential Integral functions and then look up asymptotic expansions, which give $$ I(\alpha)\sim \frac{\log\alpha}{\alpha}. $$ However, I am interested in more complicated versions these kinds of integral where the integrand might be more complicated and therefore I will need to "expand before integrating" (since I cannot evaluate the integral in closed form).

This looks a lot like a problem for Laplace's method, which finds the maximum in the exponential and expands the integrand near it, leading to a Gaussian integral. But my integrands all have the same property: the exponent is maximal not at an isolated point, but on a 1-dimensional subspace of points. Is there a theory of such integrals? I haven't found much. Any texts that mention things like "Laplace's method with a manifold of maxima" that I came across are very technical.

So my questions:

  1. Is there a straight-forward derivation of the the asymptotic expansion for the simple integral above using a Laplace-like method, i.e. without evaluating the integral in closed form to start with?

  2. Can anyone recommend not-too-technical texts that might help me?

$\endgroup$
1
$\begingroup$

You can perform a linear change of variables so that one of coordinates corresponds to the maximum of the integrand, and then integrate over this direction first ( you will be integrating a constant function, so this is easy!) After this, you will be in the setting where Laplace method will apply.

$\endgroup$

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.