Joeyonng
  • Notebook
  • Pages
  • About
  • Backyard
  1. Probability and Statistics
  2. 22  Gaussian Distribution
  • Welcome
  • Notations and Facts
  • Linear Algebra
    • 1  Fields and Spaces
    • 2  Vectors and Matrices
    • 3  Span and Linear Independence
    • 4  Basis and Dimension
    • 5  Linear Map and Rank
    • 6  Inner Product and Norm
    • 7  Orthogonality and Unitary Matrix
    • 8  Complementary Subspaces and Projection
    • 9  Orthogonal Complement and Decomposition
    • 10  SVD and Pseudoinverse
    • 11  Orthogonal and Affine Projection
    • 12  Determinants and Eigensystems
    • 13  Similarity and Diagonalization
    • 14  Normal and Hermitian Matrices
    • 15  Positive Definite Matrices
  • Calculus
    • 16  Derivatives
    • 17  Chain rule
  • Probability and Statistics
    • 18  Probability
    • 19  Random Variables
    • 20  Expectation
    • 21  Common Distributions
    • 22  Gaussian Distribution
    • 23  Moment Generating Function
    • 24  Concentration Inequalities I
    • 25  Convergence
    • 26  Limit Theorems
    • 27  Maximum Likelihood Estimation
    • 28  Bayesian Estimation
    • 29  Expectation-maximization
    • 30  Concentration Inequalities II
  • Learning Theory
    • 31  Statistical Learning
    • 32  Bayesian Classifier
    • 33  Effective Class Size
    • 34  Empirical Risk Minimization
    • 35  Uniform Convergence
    • 36  PAC Learning
    • 37  Rademacher Complexity
  • Machine Learning
    • 38  Linear Discriminant
    • 39  Perceptron
    • 40  Logistic Regression
    • 41  Multi-layer Perceptron
    • 42  Boosting
    • 43  Support Vector Machine
    • 44  Decision Tree
    • 45  Principle Component Analysis

Table of contents

  • Standard normal distribution
  1. Probability and Statistics
  2. 22  Gaussian Distribution

22  Gaussian Distribution

A random variable X \in \mathbb{R} follows the Gaussian distribution

X \sim \mathcal{N} (\mu, \sigma^{2}) \quad \mu \in \mathbb{R}, \sigma^{2} \in \mathbb{R},

if X follows the standard bell curve.

\mathbb{P}_{X} (x) = \frac{ 1 }{ \sigma \sqrt{2 \pi} } \exp \left[ - \frac{ (x - \mu)^{2} }{ 2 \sigma^{2} } \right]

\mathbb{E}_{X} [X] = \mu

\mathrm{Var} (X) = \sigma^{2}

Corollary 22.1 (Linear transformation) Let X \sim \mathcal{N} (\mu, \sigma^{2}). For any a, b \in \mathbb{R} with a \neq 0,

a X + b \sim \mathcal{N} (a \mu + b, a^{2} \sigma^{2}).

Standard normal distribution

Corollary 22.2 (Standardization) Let X \sim \mathcal{N} (\mu, \sigma^{2}). Taking a = \frac{1}{\sigma}, b = -\frac{\mu}{\sigma} in Corollary 22.1 gives the standardization of X

\frac{ X - \mu }{ \sigma } \sim \mathcal{N} (0, 1).

Definition 22.1 (Standard normal distribution) Applying the standardization to a Gaussian random variable X gives a new random variable that follows the standard normal distribution Z

Z \sim \mathcal{N} (0, 1).

Its CDF is usually denoted by \Phi

\Phi (z) = \mathbb{P} (Z \leq z), \quad z \in \mathbb{R},

which does not have a closed form.

Definition 22.2 (Standard normal quantile) For \alpha \in (0, 1), the upper \alpha quantile and the lower \alpha quantile of the standard normal distribution are the values that leave probability \alpha in the right and left tail respectively

\mathbb{P} (Z \geq z_{\alpha}) = \alpha, \qquad \mathbb{P} (Z \leq -z_{\alpha}) = \alpha, \qquad Z \sim \mathcal{N} (0, 1),

where z_{\alpha} \in \mathbb{R} is the upper \alpha quantile and -z_{\alpha} \in \mathbb{R} is the lower \alpha quantile.

Corollary 22.3 (Two-sided quantile) For \alpha \in (0, 1),

\mathbb{P} \left( -z_{\alpha / 2} \leq Z \leq z_{\alpha / 2} \right) = 1 - \alpha, \quad Z \sim \mathcal{N} (0, 1).

NoteProof

By Definition 22.2 with \alpha / 2 in place of \alpha,

\mathbb{P} (Z \geq z_{\alpha / 2}) = \frac{ \alpha }{ 2 }, \qquad \mathbb{P} (Z \leq -z_{\alpha / 2}) = \frac{ \alpha }{ 2 }.

The events \{ Z < -z_{\alpha / 2} \}, \{ -z_{\alpha / 2} \leq Z \leq z_{\alpha / 2} \}, and \{ Z > z_{\alpha / 2} \} partition the sample space, so by the countable additivity and normalization axioms in Definition 18.1,

\mathbb{P} (Z < -z_{\alpha / 2}) + \mathbb{P} \left( -z_{\alpha / 2} \leq Z \leq z_{\alpha / 2} \right) + \mathbb{P} (Z > z_{\alpha / 2}) = 1.

Since Z is continuous, \mathbb{P} (Z = z_{\alpha / 2}) = \mathbb{P} (Z = -z_{\alpha / 2}) = 0, so \mathbb{P} (Z < -z_{\alpha / 2}) = \frac{ \alpha }{ 2 } and \mathbb{P} (Z > z_{\alpha / 2}) = \frac{ \alpha }{ 2 }. Substituting and rearranging gives

\mathbb{P} \left( -z_{\alpha / 2} \leq Z \leq z_{\alpha / 2} \right) = 1 - \frac{ \alpha }{ 2 } - \frac{ \alpha }{ 2 } = 1 - \alpha.

21  Common Distributions
23  Moment Generating Function