Half-normal distribution

From Wikipedia, the free encyclopedia
Jump to: navigation, search
Half-normal distribution
Parameters σ2 > 0 — (scale)
Support x ∈ [0,\infty)
PDF f(x; \sigma) = \frac{\sqrt{2}}{\sigma\sqrt{\pi}}\exp  \left( -\frac{x^2}{2\sigma^2} \right) \quad x>0
CDF F(x; \sigma) = \mbox{erf}\left(\frac{x}{\sigma\sqrt{2}}\right)
Quantile Q(F;\sigma)=\sigma\sqrt{2}\mbox{erf}^{-1}(F)
Mean \frac{\sigma\sqrt{2}}{\sqrt{\pi}}
Median \sigma\sqrt{2}\mbox{erf}^{-1}(1/2)
Variance \sigma^2\left(1 - \frac{2}{\pi}\right)
Entropy  \frac{1}{2}  \log \left( \frac{ \pi \sigma^2 }{2} \right) + \frac{1}{2}

The half-normal distribution is a special case of the folded normal distribution.

Let X follow an ordinary normal distribution, N(0,\sigma^2), then Y=|X| follows a half-normal distribution. Thus, the half-normal distribution is a fold at the mean of an ordinary normal distribution with mean zero.

Using the \sigma parametrization of the normal distribution, the probability density function (PDF) of the half-normal is given by

f_Y(y; \sigma) = \frac{\sqrt{2}}{\sigma\sqrt{\pi}}\exp  \left( -\frac{y^2}{2\sigma^2} \right) \quad y>0,

Where E[Y] = \mu = \frac{\sigma\sqrt{2}}{\sqrt{\pi}}.

Alternatively using a scaled precision (inverse of the variance) parametrization (to avoid issues if \sigma is near zero), obtained by setting \theta=\frac{\sqrt{\pi}}{\sigma\sqrt{2}}, the probability density function is given by

f_Y(y; \theta) = \frac{2\theta}{\pi}\exp  \left( -\frac{y^2\theta^2}{\pi} \right) \quad y>0,

where E[Y] = \mu = \frac{1}{\theta}.

The cumulative distribution function (CDF) is given by

F_Y(y; \sigma) = \int_0^y \frac{1}{\sigma}\sqrt{\frac{2}{\pi}} \, \exp \left( -\frac{x^2}{2\sigma^2} \right)\, dx

Using the change-of-variables z = x/(\sqrt{2}\sigma), the CDF can be written as

F_Y(y; \sigma) = \frac{2}{\sqrt{\pi}} \,\int_0^{y/(\sqrt{2}\sigma)}\exp \left(-z^2\right)dz  = \mbox{erf}\left(\frac{y}{\sqrt{2}\sigma}\right),

where erf(x) is the error function, a standard function in many mathematical software packages.

The quantile function (or inverse CDF) is written:


where 0\le F \le 1 and \mbox{erf}^{-1}() is the inverse error function

The expectation is then given by

E(Y) = \sigma \sqrt{2/\pi},

The variance is given by

\operatorname{Var}(Y) = \sigma^2\left(1 - \frac{2}{\pi}\right).

Since this is proportional to the variance σ2 of X, σ can be seen as a scale parameter of the new distribution.

The entropy of the half-normal distribution is exactly one bit less the entropy of a zero-mean normal distribution with the same second moment about 0. This can be understood intuitively since the magnitude operator reduces information by one bit (if the probability distribution at its input is even). Alternatively, since a half-normal distribution is always positive, the one bit it would take to record whether a standard normal random variable were positive (say, a 1) or negative (say, a 0) is no longer necessary. Thus,

 H(Y) = \frac{1}{2}  \log \left( \frac{ \pi \sigma^2 }{2} \right) + \frac{1}{2}

Differential equation

\left\{\sigma ^2 f'(x)+x f(x)=0,f(1)=\frac{\sqrt{\frac{2}{\pi }}
   e^{-\frac{1}{2 \sigma ^2}}}{\sigma }\right\}

\left\{\pi  f'(x)+2 \theta ^2 x f(x)=0,f(1)=\frac{2 e^{-\frac{\theta
   ^2}{\pi }} \theta }{\pi }\right\}

Parameter estimation[edit]

Given numbers \{x_i\}_{i=1}^n drawn from a half-normal distribution, the unknown parameter \sigma of that distribution can be estimated by the method of maximum likelihood, giving

 \hat \sigma = \sqrt{\frac 1 n \sum_{i=1}^n x_i^2}

Related distributions[edit]

External links[edit]

(note that MathWorld uses the parameter  \theta = \frac{1}{\sigma}\sqrt {\pi/2} )