8
$\begingroup$

In The Fourier Transform of the quartic Gaussian $\exp(-Ax^4)$: Hypergeometric functions, power series, steepest descent asymptotics and hyperasymptotics and extensions to $\exp(-Ax^{2n})$, Boyd derives the asymptotic of the integral $$\mathcal{A}(x) = \int_{-\infty}^\infty e^{ikx} e^{-k^4}\,dk$$ for large $|x|$ using the method of steepest descent from complex analysis. The result is that $\mathcal{A}$ is a decaying oscillation with amplitude $\mathcal{O}\left(x^{-1/3}e^{-Cx^{4/3}}\right)$ for some $C>0$. Is there a way to derive this result using real-variable techniques?

$\endgroup$

1 Answer 1

12
$\begingroup$

A differential equation for ${\cal A} (x) $ can be obtained as follows, $$ \frac{d^3}{dx^3 } {\cal A} (x) = \int_{-\infty }^{\infty } dk\, (-ik^3 ) e^{ikx} e^{-k^4 } = \frac{x}{4} \int_{-\infty }^{\infty } dk\, e^{ikx} e^{-k^4 } = \frac{x}{4} {\cal A} (x) $$ where integration by parts has been used in the second equality. The leading asymptotic behavior of this differential equation is satisfied by $$ {\cal A} (x) = \exp (-C x^{4/3} ) $$ with $$ C = e^{\pm i\pi /3 } \frac{3}{4^{4/3} } = \frac{3}{4^{4/3} } \left( \frac{1}{2} \pm i \frac{\sqrt{3} }{2} \right) $$ (the third solution, with $C$ real and negative, is excluded since it diverges at large $x$). The two solutions must be suitably linearly combined to construct a real ${\cal A} (x)$. To obtain power corrections, make the ansatz $$ {\cal A} (x) = f(x) \exp (-C x^{4/3} ) $$ yielding the following differential equation for $f$, \begin{eqnarray*} & f''' - 4C x^{1/3} f'' + 3\left( C^2 \frac{16}{9} x^{2/3} - C\frac{4}{9} x^{-2/3} \right) f' & \\ & + \left( x\left[ -\frac{1}{4} -C^3 \frac{64}{27} \right] + C^2 \frac{16}{9} x^{-1/3} + C\frac{8}{27} x^{-5/3} \right) f=0 & \end{eqnarray*} The leading order (i.e., the term in the square brackets) is of course already satisfied by the above choices of $C$, but now we can also read off the next order: Namely, the terms of order $x^{-1/3} f$ and $x^{2/3} f' $ must be brought to cancel. This requires that $f$ behaves as $f\sim x^{-1/3} $ such that the derivative yields the factor $-1/3$ that is needed to effect the cancellation. Thus, also the $x^{-1/3} $ power correction is clear. Further orders can be obtained systematically. $$ $$

$\endgroup$
2
  • $\begingroup$ How did you find the leading behavior? $\endgroup$
    – Dispersion
    Nov 6, 2021 at 20:36
  • $\begingroup$ @Zachary - By inspection: Say I have $x^{\alpha } $ in the exponent, then I get a prefactor $x^{\alpha -1} $ every time I take a derivative. Note that derivatives acting on previously generated prefactors only give subleading terms. So, I need $3(\alpha -1) =1$, i.e., $\alpha = 4/3$. $\endgroup$ Nov 6, 2021 at 20:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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