Types of Distributions

For a probability space $(\Omega, \mathcal{F}, \mathbb{P})$ and measure space $(\mathbb{R}, \mathcal{B})$, let $X$ be a random variable mapping a preimage $X^{-1}(A)$ to $A \in \mathcal{B}$. Recall that $\mu(A) = \mathbb{P}(X^{-1}(A))$.

Equal in distribution

If $X$ and $Y$ induce the same distribution on $(\mathbb{R}, \mathcal{B})$, i.e., $F_X = F_Y$, then we say they have the same distribution or they are equal in distribution, denoted by:

$$X \overset{ \text{d} }{=} Y$$

Important

In general, $X(\omega) \neq Y(\omega)$ even when $X \overset{ \text{d} }{=} Y$.

Example: Consider $\Omega = \{H, T\}$ with $\mathbb{P}(H)=\mathbb{P}(T)=0.5$. Let $X(H)=1, X(T)=0$ and $Y(H)=0, Y(T)=1$. Then $X \overset{ \text{d} }{=} Y$ (both are Bernoulli(0.5)), but $X \neq Y$ everywhere.

Absolutely continuous distributions

Definition: Probability Density Function

If there exists a function $f$ such that for all $x \in \mathbb{R}$,

$$\mathbb{P}(X \le x) = F(x) = \int_{-\infty}^x f(y) \, dy$$

then $f$ is called the probability density function (PDF).

If a distribution has a density function, it is called absolutely continuous. Note that this is different from just “continuous”, which is defined below.

Properties: For any $a < b$:

$$\begin{aligned} \mathbb{P}(X \in (a, b]) &= F(b) - F(a) \\ &= \int_{-\infty}^b f(y) \, dy - \int_{-\infty}^a f(y) \, dy \\ &= \int_a^b f(y) \, dy \end{aligned}$$

Also, $\mathbb{P}(X=x) \le \mathbb{P}(X \in (x-\epsilon, x+\epsilon]) = \int_{x-\epsilon}^{x+\epsilon} f(y) \, dy$. Taking the limit as $\epsilon \to 0$, we get:

$$\implies \text{for a r.v. with a density function, } \mathbb{P}(X=x) = 0$$

This also implies $\mathbb{P}(a < X < b) = \mathbb{P}(a \le X \le b) = \int_a^b f(y) \, dy$.

Continuous distributions

Definition: Continuous Distribution

If a distribution has:

$$\mathbb{P}(X=x) = F(x) - F(x-) = 0 \quad \text{for all } x$$

i.e., its distribution function $F$ is continuous, then the distribution is called continuous.

Relationship:

• Absolutely continuous $\implies$ continuous.

• But, continuous $\;\not\!\!\!\implies$ absolutely continuous. (There exist continuous distributions which do not have a density function).

  Example: The Cantor Distribution is continuous ($F$ is continuous) but “singular” with respect to Lebesgue measure (derivative is 0 almost everywhere), so it has no PDF.

Singular distributions

A distribution is called singular if there exists a set $A \in \mathcal{B}$ such that $\lambda(A) = 0$ (Lebesgue measure is 0), but $\mathbb{P}(X \in A) = 1$, while $X$ is still continuous.

Look at the Cantor set or some fractals for examples.

Decomposition

This is why a general distribution cannot be assumed to be a mixture of just a part with density and a part with probability mass on discrete points. There can also be “singular” parts.

A general distribution can be decomposed as:

$$\text{General Distr} = \underbrace{\text{Abs. Continuous Part}}_{\text{with density } f} + \underbrace{\text{Discrete Part}}_{\text{with point masses}} + \underbrace{\text{Singular Part}}_{\text{continuous but no density}}$$

Examples

Examples

1. Uniform distribution on $[0, 1]$: $\text{Unif}[0, 1]$ or $U[0, 1]$.

   $$f(x) = 1 \quad \text{for } x \in [0, 1]$$ $$F(x) = \begin{cases} 0 & x < 0 \\ x & x \in [0, 1) \\ 1 & x \ge 1 \end{cases}$$

   

2. Exponential distribution:

   $$f(x) = \begin{cases} \lambda e^{-\lambda x} & x \ge 0 \\ 0 & \text{otherwise} \end{cases}$$

   Computation of CDF for $x \ge 0$:

   $$F(x) = \int_{-\infty}^x f(y) \, dy = \int_0^x \lambda e^{-\lambda y} \, dy = -e^{-\lambda y} \Big|_0^x = 1 - e^{-\lambda x}$$

   (and $F(x)=0$ for $x<0$).

   

3. Standard Normal: $x \in \mathbb{R}$

   $$f(x) = \frac{1}{\sqrt{2\pi}} e^{-x^2/2}$$

   The distribution function $F(x)$ does not have a closed form expression in this case.

   

4. Point mass at 0. $\mathbb{P}(X=0) = 1$.

   $$F(x) = \begin{cases} 0 & x < 0 \\ 1 & x \ge 0 \end{cases}$$