How can you derive the spacetime Fourier transform of the free Schrodinger evolution rigorously?

76 Views Asked by At

I'm trying to compute the spacetime Fourier transform of the free Schrodinger evolution. Consider $f\in L^2(\mathbb{R}^d)$ and $e^{it\Delta}f=:\mathcal{F}^{-1}(e^{-it|\xi|^2}\hat{f}(\xi))$ its free Schrodinger evolution. I wish to find its spacetime FT $\mathcal{F}_{t,x}(e^{it\Delta}f)$, defined in $\mathcal{S}'(\mathbb{R}\times\mathbb{R}^d)$ by $$ \langle \mathcal{F}_{t,x}(e^{it\Delta}f),\varphi \rangle=\langle e^{it\Delta}f,\mathcal{F}_{t,x}(\varphi) \rangle $$ for all $\varphi\in \mathcal{S}(\mathbb{R}\times\mathbb{R}^d)$, where \begin{equation} \mathcal{F}_{t,x}(\varphi)(\tau,\xi):=\int_x\int_t e^{-i(t\tau+x\cdot\xi)}\varphi(t,x)dtdx \text{ (1)} \end{equation} How can one go about this rigorously? I expect to find $$ \mathcal{F}_{t,x}(e^{it\Delta}f)(\tau,\xi)=\delta(\tau+|\xi|^2)\hat{f}(\xi) $$ interpreted as $$ \langle \delta(\tau+|\xi|^2)\hat{f}(\xi),\varphi\rangle=\int_{\{\tau=-|\xi|^2\}}\frac{\hat{f}(\xi)\varphi(\tau,\xi)}{2\sqrt{\tau^2+|\xi|^2}}d\xi d\tau $$ although I'm not completely sure if this is actually the right definition (or particularly rigorous in itself).

Part of my problem is in expressing $\mathcal{F}_{t,x}(e^{it\Delta}f)$ as an integral similar to (1) (since I am starting only from the dual definition of the FT) from which one could apply some oscillatory integral techniques.

1

There are 1 best solutions below

0
On

I think I have an answer so I will post it here lest it be helpful to anyone else:

NB: Throughout I will use the notation $\tilde{u}=\mathcal{F}_{(t,x)}u$.

Suppose first $f\in \mathcal{S}(\mathbb{R}^d)$, set $u(x,t):=e^{it\Delta}f \in L^\infty_tL^2_x$. Fix $\chi\in C_c^\infty(\mathbb{R}\times \mathbb{R}^d)$ with $\chi(0,0)=1$ and set $\chi_\epsilon(t,x):=\chi(\epsilon t,\epsilon x)$. Then \begin{align*} \langle \tilde{u},\varphi\rangle\equiv&\langle u, \tilde{\varphi}\rangle\\ =&\lim_{\epsilon\to0}\langle u,\chi_{\epsilon}\tilde{\varphi}\rangle\\ =&\lim_{\epsilon\to0}\int_t\int_x u(x,t)\chi_{\epsilon}(x,t)\left(\int_\xi\int_\tau e^{-it\tau-ix\cdot\xi}\varphi(\xi,\tau)d\xi d\tau\right)dxdt\\ =&\lim_{\epsilon\to0}\int_\tau\int_\xi \varphi(\xi,\tau) \left(\int_t\int_x e^{-it\tau-ix\cdot\xi}u(x,t)\chi_{\epsilon}(x,t)dxdt\right)d\xi d\tau\\ =&\lim_{\epsilon\to0}\langle \widetilde{u\chi_{\epsilon}},\varphi\rangle \end{align*} where we were able to use Fubini's theorem thanks to the integrability of $\varphi$, $\chi_{\epsilon}$ and $u$. Now, \begin{align*} \int_t\int_x e^{-it\tau-ix\cdot\xi}e^{it\Delta}f(x,t)\chi_{\epsilon}(x,t)dxdt=&\int_{t,x}e^{-it\tau-ix\cdot\xi}\chi_{\epsilon}(x,t)\left(\frac{1}{(2\pi)^d}\int_{\lambda\in\mathbb{R}^d}e^{ix\cdot\lambda}e^{-it|\lambda|^2}\hat{f}(\lambda)d\lambda\right)dxdt\\ =&\frac{1}{(2\pi)^d}\int_\lambda \hat{f}(\lambda)\int_{x,t}e^{-ix\cdot(\xi-\lambda)}e^{-it(\tau+|\lambda|^2)}\chi_\epsilon(x,t)dxdt d\lambda\\ =&\frac{1}{(2\pi)^d}\int_\lambda \hat{f}(\lambda)\tilde{\chi}_\epsilon(\xi-\lambda,\tau+|\lambda|^2)d\lambda \end{align*} Note that $\widetilde{\chi}_{\epsilon}(\xi,\tau)=\epsilon^{-(1+d)}\widetilde{\chi}(\epsilon^{-1}\xi,\epsilon^{-1}\tau)$, so \begin{align*} \langle \tilde{u},\varphi\rangle=&\lim_{\epsilon\to0}\int_{\tau,\xi}\varphi(\xi,\tau)\left(\frac{1}{(2\pi)^d}\int_\lambda\hat{f}(\lambda)\epsilon^{-(d+1)}\tilde{\chi}(\epsilon^{-1}(\xi-\lambda),\epsilon^{-1}(\tau+|\lambda|^2))d\lambda\right) d\xi d\tau\\ =&\lim_{\epsilon\to0}\int_\lambda\hat{f}(\lambda)\epsilon^{-(d+1)}\left(\frac{1}{(2\pi)^d}\int_{\tau,\xi}\varphi(\xi,\tau)\tilde{\chi}(\epsilon^{-1}(\xi-\lambda),\epsilon^{-1}(\tau+|\lambda|^2))d\xi d\tau\right)d\lambda\\ =&\lim_{\epsilon\to0}\int_\lambda \hat{f}(\lambda)\left(\frac{1}{(2\pi)^d}\int_{\tau',\xi'}\varphi(\epsilon \xi'+\lambda,\epsilon\tau'-|\lambda|^2)\widetilde{\chi}(\xi',\tau')d\xi'd\tau'\right)d\lambda\\ =&2\pi\int_{\lambda}\hat{f}(\lambda)\varphi(-|\lambda|^2,\lambda)d\lambda\\ \equiv&\langle 2\pi\hat{f}(\lambda)\delta(\tau+|\lambda|^2),\varphi\rangle \end{align*} where we used the dominated convergence theorem and the fact that $\int_{\tau',\xi'}\tilde{\chi}(\xi',\tau')d\xi'd\tau'=(2\pi)^{d+1}\chi(0,0)=(2\pi)^{d+1}$ to take the limit.

We can then extend this to any $f\in L^2(\mathbb{R}^d)$ be density.

Remarks:

  1. This is not quite the expected solution thanks to the factor of $2\pi$. Since the $\delta$-function actually has oscillatory integral representation $$ \delta_0=\frac{1}{(2\pi)^d}\int_{\mathbb{R}^d}e^{ix\cdot\theta}d\theta $$ I think the factor of $2\pi$ is actually correct. Nonetheless if someone sees a mistake in the above or can confirm the factor of $2\pi$ I would be grateful!

  2. Note this is not the exact same definition for the $\delta$ function as given in the question statement, however this definition seems more natural anyway, and avoids problems involving the zeros of the denominator. (For functions with non-vanishing gradient, I believe one could show the two definitions are the same by means of the co-area formula).