= Fourier-cutoff recovery of expectations from a characteristic function
Let $f$ be compactly supported and continuous, with Fourier convention $\widehat f(t)=\int f(x)e^{itx}\,dx$. Let $g,\widehat g\in L^1(\mathbb R)$, choosing $g$ as its continuous Fourier-inversion representative, and suppose $g(0)\ne0$. Then
$$
\mathbb E[f(Z)]=\lim_{\varepsilon\downarrow0}\frac1{2\pi g(0)}\int_{\mathbb R}g(\varepsilon s)\widehat f(-s)\varphi_Z(s)\,ds.
$$
<Fubini theorem> rewrites the <integral> as an average of translates of $f$ against the rescaled integrable kernel $\widehat g$, whose total mass is $2\pi g(0)$. <Dominated convergence> gives the <limit>. Requiring the continuous representative matters: arbitrarily changing the value $g(0)$ does not change its <Fourier transform> but would incorrectly change the denominator. The formula shows that a <characteristic function> determines every compactly supported continuous test expectation.
Back to article page