Consider the wave equation in one dimension $u_{tt}-u_{xx}=0$ together with a Fourier Transform along $t$, ie $$\text{FT}[u](x,\omega)=\int_{-\infty}^{+\infty}u(x,t)\exp(-i\omega t)\mathrm{d}t.\tag{1}$$ The above PDE transforms into $\partial_{xx}\text{FT}[u]+\omega^2\text{FT}[u]=0$ whose general solution reads $$\text{FT}[u](x,\omega)=A(\omega)\cos\omega x+B(\omega)\sin\omega x\tag{2}$$ which is essentially the Fourier Transform of d'Alembert's solution.

Under which conditions on $u(x,t)$ is the classical differentiation of $\text{FT}[u](x,\omega)$ with respect to $x$ meaningful? When it is meaningful, is $\partial_x \text{FT}[u]$ the Fourier Transform of $u_x(x,t)$ that is $\text{FT}[u_x]$? It is a classical result which is always used when solving PDE via Fourier Transform (and used above in the quantity $\partial_{xx} FT[u]$), however I would like to read the exact assumptions on $u$. For instance, is this differentiation acceptable when $u_{xx}(x,t)$ should be read in the sense of distributions because $u_x(x,t)$ is discontinuous?

  • 1
    $\begingroup$ A sufficient condition is the uniform convergence of the Fourier Transform of $u_{xx}(x,t)$. $\endgroup$ – Mark Viola May 25 '20 at 22:57
  • $\begingroup$ @MarkViola Thanks. Do you have a reference on this? $\endgroup$ – pluton May 26 '20 at 0:40
  • $\begingroup$ Any book on Advanced Calculus should suffice. $\endgroup$ – Mark Viola May 26 '20 at 1:17
  • $\begingroup$ @MarkViola I've edited my question with the beginning of an answer: is this what you had in mind? Thx $\endgroup$ – pluton Jun 4 '20 at 19:30
  • $\begingroup$ No. I was not extending the Fourier Transform to include tempered distributions. $\endgroup$ – Mark Viola Jun 4 '20 at 19:33

A partial answer to the above question is available in the book "Fourier Analysis, by TW Körner, Cambridge University Press, 1988, page 268, Theorem 53.5" (where $x$ and $t$ should be interchanged to comply with the question):

Let $g:\mathbb{R}\times\mathbb{R}\to\mathbb{C}$ be a continuous function such that $g_2$ exists and is continuous. Suppose $\int_{-\infty}^{+\infty}|g(x,t)|\mathrm{d}x$ and $\int_{-\infty}^{+\infty}|g_2(x,t)|\mathrm{d}x$ exist for each $t$ and that $\int_{|x|>R}|g_2(x,t)|\mathrm{d}x\to 0$ as $R\to \infty$ uniformly in $t$ on each $[a,b]$. Then $\int_{-\infty}^{+\infty}g(x,t)\mathrm{d}x$ is differentiable with $$\frac{d}{dt}\int_{-\infty}^{+\infty}g(x,t)\mathrm{d}x=\int_{-\infty}^{+\infty}\frac{\partial g}{\partial t}(x,t)\mathrm{d}x$$

[note by OP] where $g_2$ is the first partial derivative of $g$ with respect to its second argument.


If we consider for simplicity the left propagating wave, the solution reads $u(x,t)=T(x+t)$ where $T$ is a distribution. Its Fourier Transform in time is (because of translation) $$ \text{FT}[u](x,\omega)=\int_{-\infty}^{+\infty}T(x+t)\exp(-i\omega t)\mathrm{d}t=\exp(i\omega x)\text{FT}[T](\omega) \tag{3}$$ and so $\partial_x \text{FT}[u]$ and $\partial_{xx}\text{FT}[u]$ are well defined as soon as $T$ is a tempered distribution and $$\partial_x \text{FT}[u](x,\omega)=i\omega\text{FT}[u](\omega)\tag{4}$$

Let us now have a look at the Fourier Transform of $u_x=T_x$ (in the sense of distributions) $$ \begin{aligned} \text{FT}[u_x](x,\omega)&=\int_{-\infty}^{+\infty}T'(x+t)\exp(-i\omega t)\mathrm{d}t\\ &=\exp(i\omega x)\text{FT}[T'](\omega)=i\omega\exp(i\omega x)\text{FT}[T](\omega)=i\omega\text{FT}[u](x,\omega) \end{aligned} \tag{5}$$ and Equations (5) and (4) are identical.

Conclusion: for the wave equation in 1D with solution $u(x,t)=T(x+t)$, the classical differentiation with respect to space of the Fourier Transform in time is legitimate as soon as $T$ is a tempered distribution and the following holds: $$\partial_x \text{FT}[u](x,\omega)=\text{FT}[u_x](x,\omega)=i \omega \text{FT}[u](x,\omega)$$ All this is probably obvious and agrees well with (2) :). Same derivations apply for the right propagating wave $V(x-t)$.


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.