Lévy's Inversion Formula
Theorem
Let $\struct {\Omega, \Sigma, \Pr}$ be a probability space.
Let $X$ be a real-valued random variable on $\struct {\Omega, \Sigma, \Pr}$.
Let $ F_X$ be the distribution function of $X$.
Let $\phi : \R \to \C$ be the characteristic function of $X$.
Then for all $a < b$ such that:
- $\map \Pr {X \in \set {a,b} } = 0$
we have:
\(\ds \map {F_X} b - \map {F_X} a\) | \(=\) | \(\ds \map \Pr {a < X \le b}\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \lim _{T \to \infty} \frac 1 {2 \pi} \int^T _{-T} \dfrac {e^{-ita} - e^{-itb} }{it} \map \phi t \rd t\) |
Integrable Characteristic Function
Let $X$ be a real-valued random variable.
Let $P_X$ be the probability distribution of $X$.
Let $\phi : \R \to \C$ be the characteristic function of $X$.
Suppose that $\phi$ is Lebesgue integrable, i.e.:
- $\ds \int _\R \cmod {\map \phi t} \rd t < + \infty$
Then $P_X$ is absolutely continuous with respect to the Lebesgue measure.
Moreover, the Radon-Nikodym derivative is given by:
- $\ds \map g x := \dfrac 1 {2 \pi} \int_\R \map \phi t e^{- i t x} \rd t$
Proof
In fact, the first equality holds for all $a < b$, as:
\(\ds \map {F_X} b - \map {F_X} a\) | \(=\) | \(\ds \map \Pr {X \le b} - \map \Pr {X \le a}\) | Definition of $F_X$ | |||||||||||
\(\ds \) | \(=\) | \(\ds \map \Pr {\set {X \le b} \setminus \set {X \le a} }\) | Measure of Set Difference with Subset, as $\set {X \le a} \subseteq \set {X \le b}$ | |||||||||||
\(\ds \) | \(=\) | \(\ds \map \Pr {a < X \le b}\) |
In the following, we shall show the second equality.
Let $\mu$ be the probability distribution of $X$.
Let $m$ be the Lebesgue measure on $\R$.
For $T > 0$ let $m_T$ be the restriction of $m$ to $\closedint {-T} T$, i.e.:
- $\forall A \in \map \BB \R : \map {m_T} A := \map m { A \cap \closedint {-T} T}$
Let $\mu \times m_T$ be the product measure.
Then:
\(\ds \int^T _{-T} \dfrac {e^{-ita} - e^{-itb} }{it} \map \phi t \rd t\) | \(=\) | \(\ds \int^T _{-T} \int \dfrac {e^{-ita} - e^{-itb} }{it} e^{i t x} \rd \map \mu x \rd t\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \iint \dfrac {e^{it \paren {x - a} } - e^{it \paren {x - b} } }{it} \rd \map \mu x \rd \map {m_T} t\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \iint \map f {x,t} \rd \map \mu x \rd \map {m_T} t\) | $(1)$ |
where:
- $\ds \map f {x,t} := \dfrac {e^{it \paren {x - a} } - e^{it \paren {x - b} } }{it}$
The $f$ is essentially bounded with respect to $\mu \times m_T$ , since:
- $\forall \struct {x, t} \in \R \times \R_{\ne 0} : \cmod {\dfrac {e^{it \paren {x - a} } - e^{it \paren {x - b} } }{it} } \le \dfrac 3 2$
by Bounds for Complex Exponential.
In particular, $f$ is $\mu \times m_T$-integrable, as $\mu \times m_T$ is finite.
Thus by Fubini's Theorem:
\(\ds (1)\) | \(=\) | \(\ds \int f \; \rd \paren {\mu \times m_T }\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \iint \map f {x,t} \rd \map {m_T} t \rd \map \mu x\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \int F_T \rd \mu\) |
where:
\(\ds \map {F_T} x\) | \(:=\) | \(\ds \int \map f {x,t} \rd \map {m_T} t\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \int _{-T}^T \dfrac {e^{it \paren {x - a} } - e^{it \paren {x - b} } }{it} \rd t\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \int _{-T}^T \dfrac {e^{it \paren {x - a} } }{it} \rd t - \int _{-T}^T \dfrac { e^{it \paren {x - b} } }{it} \rd t\) |
Observe:
\(\ds \int _{-T}^T \dfrac {e^{it \paren {x - a} } }{it} \rd t\) | \(=\) | \(\ds \int _{-T}^T \dfrac {\map \cos {t \paren {x-a} } + i \map \sin {t \paren {x-a} } }{it} \rd t\) | Euler's Formula | |||||||||||
\(\ds \) | \(=\) | \(\ds \int _{-T}^T \dfrac {\map \sin {t \paren {x-a} } }{t} \rd t\) | Definite Integral of Odd Function | |||||||||||
\(\ds \) | \(=\) | \(\ds 2 \int _0^T \dfrac {\map \sin {t \paren {x - a} } } t \rd t\) | Definite Integral of Even Function | |||||||||||
\(\ds \) | \(=\) | \(\ds 2 \; \map \sgn {x - a} \int _0^T \dfrac {\map \sin {t \size {x - a} } } t \rd t\) | Sine Function is Odd | |||||||||||
\(\ds \) | \(=\) | \(\ds 2 \; \map \sgn {x - a} \int _0^{T \size {x-a} } \dfrac {\sin s } s \rd s\) | Integration by Substitution with $s = t \size {x-a}$ | |||||||||||
\(\ds \) | \(=\) | \(\ds 2 \; \map \sgn {x - a} \; \map \Si {T \size {x - a} }\) |
where:
- $\sgn : \R \to \set {-1, 0, 1}$ is the signum function
- $\Si : \R \to \R$ is the sine integral function
Similarly:
- $\ds \int _{-T}^T \dfrac {e^{it \paren {x - b} } }{it} \rd t = 2 \; \map \sgn {x - b} \; \map \Si {T \size {x - b} }$
Thus we have:
- $\ds \map {F_T} x = 2 \; \map \sgn {x - a} \; \map \Si {T \size {x - a} } - 2 \; \map \sgn {x - b} \; \map \Si {T \size {x - b} }$
By Limit at Infinity of Sine Integral Function, for all $x \in \R \setminus \set {a,b}$:
\(\ds \lim _{T \mathop \to +\infty} \map {F_T} x\) | \(=\) | \(\ds \pi \paren {\map \sgn {x - a} - \map \sgn {x - b} }\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds 2 \pi \; \map {\chi _{\hointl a b} } x\) |
As $\map \mu {\set {a, b} } = 0$ by hypothesis, this means in particular:
- $\ds \lim _{T \mathop \to +\infty} F_T = \chi _{\hointl a b}\quad$ $\mu$-almost surely
On the other hand, Sine Integral Function is Bounded:
- $\ds \sup_{T > 0, \; x \in \R} \size {\map {F_T} x} \le 4 \norm \Si_\infty < +\infty$
Note that the constant function $4 \norm \Si_\infty$ is $\mu$-integrable.
Therefore:
\(\ds \lim _{T \mathop \to +\infty} \int F_T \rd \mu\) | \(=\) | \(\ds \int \lim _{T \mathop \to +\infty} F_T \rd \mu\) | Lebesgue's Dominated Convergence Theorem | |||||||||||
\(\ds \) | \(=\) | \(\ds 2 \pi \int \chi _{\hointl a b} \rd\mu\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds 2 \pi \; \map \mu {\hointl a b}\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds 2 \pi \; \map \Pr {X \in \hointl a b}\) | Definition of Probability Distribution | |||||||||||
\(\ds \) | \(=\) | \(\ds 2 \pi \; \map \Pr {a < X \le b}\) |
This needs considerable tedious hard slog to complete it. In particular: Some details To discuss this page in more detail, feel free to use the talk page. When this work has been completed, you may remove this instance of {{Finish}} from the code.If you would welcome a second opinion as to whether your work is correct, add a call to {{Proofread}} the page. |
Source of Name
This entry was named for Paul Pierre Lévy.
Sources
- 1995: Patrick Billingsley: Probability and Measure (3rd ed.): $26$: Characteristic Functions