This question is about heuristics to understand with the tempered distribution given by the oscillatory integral $\int xe^{ix\xi} d\xi$ (all integrals are over $\mathbb R$). Source is M. Zworski's Semiclassical Analysis book, page 54.

Maybe it bears mentioning: the author is "not a crank" and this was supposed to be a simple example as more calculations are yet to come in later chapters. I am aware how to compute things rigourously, but it seems that the author has a way to guess the right answer beforehand that I cannot understand.

PS $\int x e^{ix\xi} d\xi= x\int e^{ix\xi} d\xi = 2\pi x\delta = 0$ as tempered distributions (obviously just heuristically but do it with Fourier inversion starting from \eqref{B} below). In contrast with the previous form of reuns' answer below which computed $\int i\xi e^{ix\xi}d\xi =D_x\int e^{ix\xi}d\xi =2\pi \delta'$.

First, its alleged that "integration by parts" (quotation marks are from the book) shows $$ \int x e^{ix\xi} d\xi = \int D_\xi(e^{ix\xi})d\xi = 0.$$ (Nevermind the missing $-i$ since things should be zero) I suppose the author intended me to envision the following? $$ \int D_\xi (e^{ix\xi}) d\xi = e^{ix\xi}|_{-\infty}^{+\infty} -\int(D_\xi1)e^{ix\xi}d\xi\tag{IBP}\label{IBP}$$ but I don't have the 'feeling' that this 'should' evaluate to zero.

Second, the author tries to justify this as follows: for some large $k$, $$ u(\psi) := \lim_{\epsilon\to 0+} \iint xe^{ix\xi} \psi(x) \chi(\epsilon\xi) dxd\xi = \iint \frac x{(1+\xi^2)^k} e^{ix\xi} (1-\partial_x^2)^k \psi(x) dxd\xi\tag{A}\label{A}$$ From here, it is said that integration by parts in $\xi$ gives the result. But I think that there's an error here and I cannot commute $x$ with the $x$ derivatives, $[(1-\partial_x^2), x]\psi = 2\psi'$? Instead what I get is either $$ u(\psi) = \iint \frac 1{(1+\xi^2)^k} e^{ix\xi} (1-\partial_x^2)^k \Big(x\psi(x)\Big) dxd\xi\tag{B}\label{B}$$ or $$ u(\psi) = \iint \frac x{(1+\xi^2)^k} e^{ix\xi} L\psi(x) dxd\xi\tag{C}\label{C}$$ where $L\neq(1-\partial_x^2)^k$ is some other differential operator. From \eqref{B}, I can conclude with the Fourier inversion formula, but I see no integration by parts. Whereas if I start from \eqref{C} and try to follow the steps in \eqref{IBP}, I'm led to $$ \int L\psi(x) \left(\int (1+\xi^2)^{-k} D_\xi e^{ix\xi}d\xi \right) dx = \int L\psi(x) \left( \left.\frac{e^{ix\xi}}{(1+\xi^2)^k}\right|_{\xi=-\infty}^\infty - \int D_\xi((1+\xi^2)^{-k}) e^{ix\xi} d\xi\right)d x$$ and while there is now decay to say $\left.\dfrac{e^{ix\xi}}{(1+\xi^2)^k}\right|_{\xi=-\infty}^\infty=0$, its not immediately obvious to me that the other term should be zero.


  • Should \eqref{IBP} be intuitive? If it should, how far is this intuition correct, is it e.g. correct only if I see power law growth in $\xi$?

  • Did I make a mistake?

  • What is the intended calculation?

  • $\begingroup$ I don’t want to bump the question for a minor edit but there is no missing $-i$ because in this book $D:=\partial/i$ $\endgroup$ – Calvin Khor Mar 7 at 5:12

There is no heuristic and you should not try to "evaluate" divergent integrals that converge only in the sense of distributions.

$$h_A(x)=x\int_{-A}^A e^{ixt}dt =2\sin(A x)$$ For $\phi \in C^\infty_c$ $$\lim_{A\to \infty}\int_{-\infty}^\infty h_A(x)\phi(x)dx= \lim_{A\to \infty}\int_{-\infty}^\infty \frac{2\cos(Ax)}{A} \phi'(x)dx=0$$ ie. $$\lim_{A\to \infty} h_A = 0 \text{ in the sense of distributions}$$

  • $\begingroup$ I agree that it’s dodgy but I still want to know what point that the author is making. Also I believe you misread, it is indeed zero because i wrote a $d\xi$ integral so it eventually simplifies to $x\delta=0$? $\endgroup$ – Calvin Khor Mar 5 at 23:22
  • $\begingroup$ Yes but it doesn't change the idea, forget about dodgy calculations, there are no such, if the book says it does then pick another one. $\endgroup$ – reuns Mar 6 at 10:25
  • $\begingroup$ I do not plan to rely on it, or necessarily even use it, but I would like to know what the author wanted to say and why he thinks it is useful. $\endgroup$ – Calvin Khor Mar 6 at 10:46
  • $\begingroup$ Nothing, people are just wrong when they are trying to give heuristics on distributions. $\endgroup$ – reuns Mar 6 at 10:49
  • $\begingroup$ See how my answer is trivial (once the definition of the distribution $x\int_{-\infty}^\infty e^{ixt}dt$ is known, ie. $\lim h_A$), zero reason to ask for an heuristic. $\endgroup$ – reuns Mar 6 at 10:52

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.