8
$\begingroup$

In the paper "The Inverse Function Theorem of Nash and Moser" by Richard S. Hamilton it is claimed that there exists a function $\phi$ such that:

$$\int_{0}^{\infty}t^{n}\phi(t)dt=(-1)^{n}$$

For $n=0,1,2,...$. In fact one is provided. The example is:

$$\phi(t)=\frac{e^{2\sqrt{2}}}{\pi(1+t)}e^{-(t^\frac{1}{4}+t^\frac{-1}{4})}\sin(t^\frac{1}{4}-t^{\frac{-1}{4}})$$

My trouble is that the paper said that verifying the integral can be done by methods of contour integration by recognizing this function as the real part of a holomorphic function. The trouble I'm having is that I have no idea how to identify this holomorphic function. Any help is greatly appreciated. Thank you in advance.

$\endgroup$
  • 1
    $\begingroup$ This reminds me of the function used in E.M. Stein's book on singular integrals being $\psi(t) = \frac{e}{\pi t} e^{-\sqrt{2}/2 (t-1)^{1/4}} sin(\sqrt{2}/2 (t-1)^{1/4}$, which satisfies $\int_1^\infty \psi(t) dt = 1$ and all higher moments vanish. $\endgroup$ – andre Aug 4 '14 at 22:03
  • 1
    $\begingroup$ $\psi(t) = \frac{e}{\pi t} e^{-\sqrt{2}/2 (t-1)^{1/4}} sin(\sqrt{2}/2 (t-1)^{1/4})$ $\endgroup$ – andre Aug 4 '14 at 22:11
  • 1
    $\begingroup$ And also the one in D.V. Widder's book on the laplace transform, being $\phi(t) = e^{-t^{1/4}}sin(t^{1/4})$, where all the moments vanish for the integral on $[0,\infty)$, see archive.org/stream/laplacetransform031816mbp#page/n141/mode/2up $\endgroup$ – andre Aug 12 '14 at 20:47
  • $\begingroup$ And as stated in D.V. Widders book the last $\phi$ can be found in a 1894 paper of Stieltjes on page J. 105, see archive.numdam.org/ARCHIVE/AFST/AFST_1894_1_8_4/…, I wonder how Stieltjes found this function. $\endgroup$ – andre Aug 13 '14 at 22:53
7
$\begingroup$

Wahoo, this is cool. Replace $t$ with $u^4$ in order to have: $$I_n=\int_{0}^{+\infty}t^n \phi(t)\,dt = \frac{4e^{2\sqrt{2}}}{\pi}\int_{0}^{+\infty}\frac{u^{4n+3}}{1+u^4}\sin(u-1/u)\exp(-u-1/u)\,du,$$ then split $[0,+\infty)=[0,1]\cup[1,+\infty)$ and use the substitution $u\leftarrow 1/u$ on the second interval in order to have: $$I_n = \frac{4e^{2\sqrt{2}}}{\pi}\int_{0}^{1}\frac{u^{4n+2}-u^{-4n-2}}{u^2+u^{-2}}\sin(u-1/u)\exp(-u-1/u)\frac{du}{u}.$$ Now substitute $u=e^{-v}$ in order to have: $$I_n = \frac{4e^{2\sqrt{2}}}{\pi}\int_{0}^{+\infty}\frac{\sinh((4n+2)v)}{\cosh(2v)}\sin(2\sinh v)\exp(-2\cosh v)\,dv,$$ and since the integrand function is even: $$I_n = \frac{2e^{2\sqrt{2}}}{\pi}\Im\int_{\mathbb{R}}\frac{\sinh((4n+2)v)}{\cosh(2v)}\exp(2i\sinh v-2\cosh v)\,dv.$$ We can approach the last integral with the usual complex analytic techniques, by integrating the function over a rectangle having vertices in $-R,R,R+\frac{\pi}{2}i,-R+\frac{\pi}{2}i$. The only singularity that matters is the one in $v=\frac{\pi}{4}i$: since the residue of the integrand function in such a point is: $$\frac{(-1)^n}{2}e^{-2\sqrt{2}}$$ we have $I_n=(-1)^n$ just as claimed.

$\endgroup$
0
$\begingroup$

In general, an analytic function must have harmonic real and imaginary parts (i.e. they satisfy Laplace's equation) and satisfy the Cauchy-Riemann equations.

In this case, replace the sine into a complex exponential.

$\endgroup$

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.