4

This is Exercise 2.4 in Stein & Shakarchi's Complex Analysis. Prove that for all $\xi \in \mathbb{C}$ $$e^{-\pi\xi^2}=\int_{-\infty}^\infty \! e^{-\pi x^2}e^{2\pi ix\xi}\ \mathrm{d}x.$$ They prove it for the real case, so I assume that I'm supposed to use that. All I can think to do is write $\xi=a+ib$, which gives a term $e^{2\pi i xa}$ in the integral which becomes 1, since the power is a real multiple of $2\pi$ and $i$. That leaves $$e^{-\pi\xi^2}=\int_{-\infty}^\infty \! e^{-\pi x^2}e^{-2\pi xb}\ \mathrm{d}x$$ but I don't have any idea where to go from there, or if this is even a fruitful direction. Any hints would be appreciated.

desi
  • 382
  • 3
    $e^{2\pi ixa}$ doesn't equal $1$ unless $xa$ is an integer! – TonyK Sep 22 '14 at 06:02
  • 4
    Hint: Complete the square in the exponential: $e^{-\pi x^2 + 2\pi i x \zeta} = e^{-\pi(x - i\zeta)^2 + \pi\zeta^2}$ and change variables to $y = x - i\zeta$. – Winther Sep 22 '14 at 06:11
  • 1
    ... using the Cauchy Integral Theorem with estimates on integrals over two line segments to justify this complex change of variables. – Robert Israel Sep 22 '14 at 06:43
  • You may show that the map $G(z)=\int_{\mathbb{R}}e^{-\pi x^2}e^{2\pi I z},dx$ is analytic. As it coincides with $F(z)=e^{-\pi z^2}$, which is analytic too, on $\mathbb{R}$, the result follows. – Mittens Apr 06 '22 at 14:14

3 Answers3

8

Write $\xi=a+bi$. The exponent inside the integrand looks like

$$\begin{array}{ll} -\pi x^2+2\pi ix\xi & =-\pi x^2+2\pi ix(a+bi) \\ & =-\pi x^2+2\pi iax-2\pi bx \\ & = -\pi (x^2+2bx)+2\pi iax \end{array}$$

The real part has both an $x$ and an $x^2$ term, but we may complete the square:

$$ \begin{array}{l} = -\pi(x^2+2bx+b^2)+\pi b^2+2\pi iax \\ =-\pi(x+b)^2+\pi b^2+2\pi iax. \end{array} $$

With the substitution $u=x+b$ ($\Leftrightarrow x=u-b$) this becomes

$$ \begin{array}{l} =-\pi u^2+\pi b^2+2\pi ia(u-b) \\ =-\pi u^2+2\pi iau+[\pi b^2-2\pi iab]. \end{array}$$

Therefore

$$ \begin{array}{ll} \displaystyle \int_{-\infty}^{\infty} \exp(-\pi x^2+2\pi ix\xi)\,\mathrm{d}x & \displaystyle =\exp(\pi b^2-2\pi iab)\int_{-\infty}^{\infty} \exp(-\pi u^2+2\pi iau)\,\mathrm{d}u \\ & \displaystyle =\exp(\pi b^2-2\pi iab)\exp(-\pi a^2) \\ & =\exp(-\pi(a+bi)^2) \\ & =\exp(-\pi \xi^2). \end{array} $$

anon
  • 151,657
  • Thanks, this cleared some confusion. Though simple, I was lost in the maniplulations and when the apply the identity for the real case.I'm not sure, why someone voted it down. – user135520 Jan 05 '17 at 04:01
  • @user135520 This is a complex analysis question, so we use the Cauchy integral theorem, or analytic continuation, or the identity theorem for analytic functions, but in any case the answer isn't given here. In the answer above it is even worse, the main argument (change of variable $u = x+i\xi$) being wrong. – reuns Jan 05 '17 at 04:55
  • @user1952009 To me the other answer is below, not above, so I thought you were saying my change of variables was wrong. In any case, you say "the answer isn't given here" - do you believe my answer is invalid? I would appreciate bringing errors to light so they do not fester. – anon Jan 05 '17 at 05:11
  • 1
    @arctictern One possible proof is $ \int_{-\infty}^\infty e^{-x^2} e^{2ax} dx= e^{a^2} \int_{-\infty}^\infty e^{-(x-a)^2}dx=\sqrt{\pi} e^{a^2} $ for every $a\in \mathbb{R}$, and since both side are complex analytic in $a$, by analytic continuation or by the identity theorem $\int_{-\infty}^\infty e^{-x^2} e^{2ax} dx = \sqrt{\pi} e^{a^2} $ stays true for every $a \in \mathbb{C}$. The other is with the Cauchy integral formula, applied to $\int_{a-\infty}^{a+\infty} e^{-(x-a)^2}dx$ – reuns Jan 05 '17 at 05:42
  • @user1952009 I was not asking for more proofs, I was asking for the error in mine. – anon Jan 05 '17 at 05:53
  • Maybe the change of variables just needs some justification? – user135520 Jan 05 '17 at 16:48
  • 2
    Well, $u(x)=x+b$ (where $b$ is real) is a smooth bijection of the domain $\mathbb{R}$ satisfying $du/dx=1$, so $\int_{-\infty}^{\infty} f(x+b)dx=\int_{-\infty}^{\infty} f(u)du$ for any function $f$ (even complex-valued, since we can split into real and imaginary parts) as long as $\int_{-\infty}^{\infty}f(u)du$ is defined unambiguously. The complex change of variables in the other answer would require more justification since it changes the domain of integration. – anon Jan 05 '17 at 17:09
2

Note that the integral $$ \int_{-\infty}^{\infty} e^{-\pi x^2}e^{-2\pi i x\xi}\,\mathrm{d}x $$ is the Fourier transform of $e^{-\pi x^2}$. Now simplify $$ e^{-\pi x^2}e^{-2\pi ix\xi} = e^{-\pi(x^2+2\pi i x)} $$ and complete the square of $x^2 + 2\pi i x$ to deduce that $$ x^2 + 2\pi i x = x^2 + 2\pi i x -\xi^2 + \xi^2 = (x+i\xi) + \xi^2. $$ Substituting, this gives $$ e^{-\pi(x^2 + 2\pi i x)} = e^{-\pi(x+i\xi)^2-\pi\xi^2} = e^{-\pi(x+i\xi)^2} e^{- \pi\xi^2} $$ and hence $$ \int_{-\infty}^{\infty} e^{-\pi(x^2+2\pi ix\xi)}\,\mathrm{d}x= e^{-\pi\xi^2}\int_{-\infty}^{\infty} e^{-\pi(x+i\xi)^2}\,\mathrm{d}x. $$ Making the substitution $u = x+i\xi$ so that $\mathrm{d}x = \mathrm{d}u$ then yields $$ e^{-\pi\xi^2}\int_{-\infty}^{\infty} e^{-\pi(x+i\xi)^2}\,\mathrm{d}x = e^{-\pi\xi^2}\int_{-\infty}^{\infty}e^{-\pi u^2}\,\mathrm{d}u= e^{-\pi\xi^2}\frac{\sqrt{\pi}}{\sqrt{\pi}} = e^{-\pi\xi^2}, $$ as was to be shown. Note that this shows that $$ \mathscr{F}(e^{-\pi x^2}) = e^{-\pi \xi^2}, $$ where $\mathscr{F}$ is the Fourier transform on $L^2(\mathbb{R})$.

Geoff
  • 1,594
2

This is, in my opinion, the most conventional solution that uses Cauchy's theorem. Hope this helps others like myself :)

Let $\gamma$ be a closed parallelogram, starting from the origin, consisting of 4 segments $\gamma_1, \gamma_2, \gamma_3, \gamma_4$ in clockwise orientation sequentially.

plotting of the curve $\gamma$

As a complex function $f(z) = e^{\pi z^2}$ is holomorphic in this simply connected region, we can apply Cauchy's theorem to get \begin{align*} 0 = \int_\gamma f(z)\, dz = \underset{(1)}{\underbrace{\int_{\gamma_1} e^{\pi z^2}\, dz}} + \underset{(2)}{\underbrace{\int_{\gamma_2} e^{\pi z^2}\, dz}} + \underset{(3)}{\underbrace{\int_{\gamma_3} e^{\pi z^2}\, dz}} + \underset{(4)}{\underbrace{\int_{\gamma_4} e^{\pi z^2}\, dz}} \end{align*}

For $(1)$, \begin{align*} \lim_{R \rightarrow \infty} (1) = \lim_{R \rightarrow \infty} \int_{-R}^R e^{\pi(ix)^2} i\, dx = i \int_{-\infty}^\infty e^{-\pi x^2} i\, dx = i \end{align*}

Assume $\xi = a + ib$, $(2)$ is bounded absolutely by \begin{align*} |(2)| &= \left| \int_{\gamma_2} e^{\pi z^2}\, dz \right| \\ &\leq \sup_{0 \leq t \leq 1} \left| exp(\pi (\xi t + iR)^2)\, dz \right| \cdot 1 \\ &=\sup_{0 \leq t \leq 1} exp(\pi ((ta)^2-(tb+R)^2))\, dz \\ \end{align*} which goes to $0$ as $R \rightarrow \infty$, so that $\lim_{R \rightarrow \infty} (2) = 0$. Similarly, $\lim_{R \rightarrow \infty} (4) = 0$.

For $(3)$, it can be written as \begin{align*} (3) = \int_R^{-R} exp(\pi(\xi + iy)^2) i\, dy = -i\int_{-R}^R exp(\pi\xi^2 -\pi y^2 + 2\pi i\xi y)\, dy = - i e^{\pi\xi^2} \int_{-R}^R e^{-\pi y^2} e^{2 \pi i \xi y}\, dy \end{align*}

Assembling $(1), (2), (3)$ gives, \begin{align*} i = i e^{\pi\xi^2} \int_{-R}^R e^{-\pi y^2} e^{2 \pi i \xi y}\, dy \end{align*} Dividing both sides with ${i e^{\pi\xi^2}}$ gives the desired answer.

bivoje
  • 21