1: \begin{proof}

2: We will sketch the main idea of the proof, and leave it to the reader to make the argument rigorous.

3: 

4: By analogy with an orthonormal expansion in the discrete case, we assume that $f \in L^2(\R)$ has a decomposition in terms of the eigenfunction $\sech x$ and the generalized eigenfunctions $L^+ v_\omega$ in the form

5: \[

6: f = c \la f , \sech \ra \sech + \int_\R m_f(\omega) \la f , L^+ v_\omega \ra L^+ v_\omega \, d\omega ,

7: \]

8: where the coefficient $c$ and multiplier $m_f(\omega)$ are to be determined.

9: 

10: Taking the inner product with $\sech x$ implies that $c=\tfrac{1}{2}$, since $\lv \sech \rv_{L^2(\R)}^2 = 2$ and $\la L^+ v_\omega , \sech \ra = \la v_\omega , L^- \sech \ra = 0$.

11: 

12: Next we annihilate the $\sech$ term by applying $L^-$ to both sides:

13: \[

14: L^- f = L^- \Big( \int_\R m_f(\omega) \la f , L^+ v_\omega \ra L^+ v_\omega \, d\omega \Big) .

15: \]

16: Note that by integration by parts,

17: \[

18: \la f , L^+ v_\omega \ra = \la L^- f , v_\omega \ra = \widehat{(L^- f)}(\omega) .

19: \]

20: Hence

21: %

22: \begin{align*}

23: L^- f

24: & = L^- \Big( \int_\R m_f(\omega) \widehat{(L^- f)}(\omega) L^+ v_\omega \, d\omega \Big) \\

25: & = \int_\R m_f(\omega) \widehat{(L^- f)}(\omega) L^- L^+ v_\omega \, d\omega \\

26: & = \int_\R m_f(\omega) \widehat{(L^- f)}(\omega) (1+4\pi^2 \omega^2) v_\omega \, d\omega

27: \end{align*}

28: %

29: by \eqref{eq:reversederiv}. Thus the multiplier should be $m_f(\omega) =1/(1+4\pi^2 \omega^2)$, in order for Fourier inversion to hold. This argument shows the necessity of the formula in the theorem, and one can show sufficiency by suitably reversing the steps.

30: \end{proof}

31: