Given the characteristic function of a continuous real random variable, it is possible to construct a finite Fourier series that approximates the density over any specified finite interval. This series converges in mean square if the density decays faster than |x|-1 and the characteristic function decays faster than |t|-1/2; it converges in minimax if the density decays faster than |x|-1 and the characteristic function decays faster than |t|-1. It is also possible to construct a finite Fourier series for the distribution function that has arbitrarily small maximum error over any finite interval provided that the characteristic function decays as $|t|^{-\beta}$ for some $\beta>0$ and that either the distribution or the density decays (to 0 or 1, as appropriate) as $|x|^{-\alpha}$ for some $\alpha>1$; these conditions are weak enough to include most continuous distributions of practical interest. Explicit error bounds can be obtained directly from the characteristic function, without prior knowledge of either the distribution or density, making these series useful for numerical inversion.
[1]
J. Gil-Pelaez.
Note on the inversion theorem
,
1951
.
[2]
S. Lang,et al.
An Introduction to Fourier Analysis and Generalised Functions
,
1959
.
[3]
J. Cooley,et al.
Application of the fast Fourier transform to computation of Fourier integrals, Fourier series, and convolution integrals
,
1967,
IEEE Transactions on Audio and Electroacoustics.
[4]
H. Bohman,et al.
A method to calculate the distribution function when the characteristic function is known
,
1970
.
[5]
R. Davies.
Numerical inversion of a characteristic function
,
1973
.
[6]
B. Schorr,et al.
Numerical inversion of a class of characteristic functions
,
1973
.
[7]
M. Kendall,et al.
Kendall's advanced theory of statistics
,
1995
.
[8]
Bruce W. Turnbull,et al.
Obtaining Distribution Functions by Numerical Inversion of Characteristic Functions with Applications
,
1995
.