1 Definition and basic properties

The Cauchy distribution is a continuous probability distribution on the real line with a distinctive bell-like shape and exceptionally heavy tails. It is often used as a standard example in probability theory because several familiar results for averages, moments, and convergence do not apply to it. In contrast with the normal distribution, the Cauchy law does not have a finite mean or variance.

1.1 Probability density function

A Cauchy distribution is commonly described by a location parameter and a scale parameter. Its probability density function is

\[ f(x)=\frac{1}{\pi \gamma \left[1+\left(\frac{x-x_0}{\gamma}\right)^2\right]}, \]

where \(x_0\) is the location and \(\gamma>0\) is the scale. The density is highest at \(x=x_0\) and declines slowly as \(x\) increases.

1.2 Cumulative distribution function

The cumulative distribution function is

\[ F(x)=\frac{1}{\pi}\arctan\!\left(\frac{x-x_0}{\gamma}\right)+\frac{1}{2}. \]

This formula shows that the distribution accumulates probability gradually in the tails, unlike distributions with exponentially decaying tails.

1.3 Support and parameterization

The support of the Cauchy distribution is the entire real line. It is usually parameterized by a location parameter, which shifts the center, and a scale parameter, which controls the spread. Several equivalent parameterizations appear in the literature, but they describe the same family of distributions.

1.4 Symmetry and median

The distribution is symmetric about its location parameter. As a result, the location parameter is also the median and the mode. This symmetry is one reason the Cauchy law is convenient in theoretical work, even though it is difficult to summarize by averages.

1.5 Heavy-tailed behavior

The tails of the Cauchy distribution decrease proportionally to \(1/x^2\) for large \(x\). This slow decay means extreme values occur much more often than they would under a normal distribution. The heavy tails are responsible for the failure of many standard limit theorems and summary statistics.

2 Standard Cauchy distribution

The standard Cauchy distribution is the best-known member of the family. It has location \(0\) and scale \(1\), and its simple form is often used as a reference case in mathematics and statistics.

2.1 Canonical form

For the standard Cauchy distribution, the density becomes

\[ f(x)=\frac{1}{\pi(1+x^2)}. \]

Its cumulative distribution function is

\[ F(x)=\frac{1}{\pi}\arctan(x)+\frac{1}{2}. \]

2.2 Scaling and location parameters

Any Cauchy distribution can be obtained from the standard form by shifting and rescaling. If \(X\) is standard Cauchy, then \(x_0+\gamma X\) has a Cauchy distribution with location \(x_0\) and scale \(\gamma\). This simple transformation property makes the standard distribution a natural base case.

2.3 Graphical interpretation

The graph of the density is centered at the location parameter and falls off slowly on both sides. Compared with the normal curve, it has a lower central peak and much thicker tails. This visual contrast is often used to illustrate the meaning of heavy-tailed behavior.

3 Characterizations

The Cauchy distribution appears in several different mathematical constructions. These characterizations help explain why the same distribution arises in geometry, trigonometry, and random-variable ratios.

3.1 Ratio of normal variables

If \(U\) and \(V\) are independent standard normal random variables, then the ratio \(U/V\) has a standard Cauchy distribution. This result is one of the most important classical characterizations of the law. It also shows how a ratio can produce a distribution with no finite mean.

3.2 Tangent transformation of a uniform variable

If \(\Theta\) is uniformly distributed on \((-\pi/2,\pi/2)\), then \(\tan(\Theta)\) is standard Cauchy. This representation connects the distribution to angular geometry and explains the appearance of the arctangent in its cumulative distribution function.

3.3 Stable distribution property

The Cauchy distribution is a stable distribution. This means that suitable linear combinations of independent Cauchy variables remain Cauchy distributed, up to changes in location and scale. Stability helps distinguish it from many common distributions.

3.3.1 Closure under addition and scaling

If independent Cauchy random variables are added, the result is again Cauchy, after appropriate rescaling. For the standard case, the sum of two independent standard Cauchy variables has the same type of law, with an adjusted scale. This closure property is one reason the distribution is central in stable-law theory.

3.4 Relation to Student's t-distribution

The Cauchy distribution is a special case of the Student's t-distribution with one degree of freedom. This relationship places it within a broader family of heavy-tailed distributions that are important in inferential statistics. As the degrees of freedom increase, the Student's t-distribution becomes closer to the normal distribution, but the one-degree-of-freedom case remains especially heavy-tailed.

4 Moments and measures of central tendency

The Cauchy distribution is notable for the failure of several usual summary measures. These failures are not technical curiosities; they are a direct consequence of the tail behavior.

4.1 Nonexistence of mean

The expected value of a Cauchy random variable does not exist in the usual sense. The positive and negative tail contributions do not combine to produce a finite integral. This makes the distribution a standard example where the average is not defined.

4.2 Nonexistence of variance

The variance is also undefined because even the first absolute moment fails to converge. Since variance depends on the second moment, the divergence is even more pronounced. This is another way in which the Cauchy law differs sharply from the normal distribution.

4.3 Median and mode

Although the mean does not exist, the median and mode are well defined and equal to the location parameter. These measures are stable under the symmetry of the distribution. In practice, they are more informative than the mean for Cauchy data.

4.4 Quantiles

Quantiles are easy to compute from the inverse of the cumulative distribution function. The \(p\)-th quantile of the standard Cauchy distribution is \(\tan[\pi(p-1/2)]\). Quantiles are often preferred when describing Cauchy data because they remain meaningful even when moments do not.

5 Functions associated with the distribution

Several analytic functions are associated with the Cauchy distribution and are useful in theory and applications. Some of these functions exist in a limited sense only, reflecting the distribution’s heavy tails.

5.1 Characteristic function

The characteristic function of a Cauchy distribution with location \(x_0\) and scale \(\gamma\) is

\[

\phi(t)=\exp(i x_0 t-\gammat).

\]

This simple form is characteristic of stable laws and shows the linear effect of location and the exponential damping from scale.

5.2 Moment generating function

The moment generating function does not exist for nonzero arguments, because the required expectations diverge. This failure is consistent with the absence of ordinary moments. As a result, standard moment-based methods are not available.

5.3 Tail distribution

The survival function decreases slowly and is comparable to a reciprocal quadratic in the tails. This behavior gives the distribution a significant probability of extreme observations. Tail probabilities are often more useful than moments when studying the Cauchy law.

5.4 Entropy

The differential entropy of the Cauchy distribution is finite. For the standard form, it equals \(\log(4\pi)\). This is a reminder that a distribution can have finite entropy even when its mean and variance are undefined.

The Cauchy distribution interacts neatly with several transformations. These related forms appear in both theoretical calculations and modeling contexts.

6.1 Affine transformations

Any affine transformation of a Cauchy random variable is again Cauchy. If \(X\) is Cauchy, then \(aX+b\) is Cauchy for any nonzero \(a\). The new location and scale are determined directly by \(a\) and \(b\).

6.2 Reciprocal distribution

The reciprocal of a standard Cauchy random variable is also standard Cauchy. This self-reciprocal property is unusual and reflects the symmetry of the density under inversion. It provides another distinctive example of the distribution’s invariance.

6.3 Log-Cauchy distribution

If a random variable is the logarithm of a positive Cauchy-related variable, the resulting law is often called log-Cauchy. Such distributions are used in situations where multiplicative variability and heavy tails are both relevant. They inherit some of the tail irregularity of the original Cauchy family.

6.4 Multivariate Cauchy distribution

There is also a multivariate Cauchy distribution, which generalizes the univariate case to several dimensions. It is associated with multivariate stable laws and can be described through suitable projection properties. Each one-dimensional projection has a Cauchy distribution.

7 Statistical inference

Inference for Cauchy data is more difficult than for light-tailed distributions. Standard estimators can behave poorly, so methods that rely on robustness or quantiles are often preferred.

7.1 Parameter estimation

Estimating the location and scale parameters of a Cauchy distribution is challenging because ordinary averages and sample variances are not reliable. Successful procedures usually exploit medians, likelihood methods, or robust loss functions.

7.1.1 Maximum likelihood estimation

Maximum likelihood estimation can be applied, but the likelihood surface may be irregular and can contain multiple local maxima. Numerical optimization is often required. Because of the heavy tails, the sample likelihood may be sensitive to starting values and outliers.

7.1.2 Robust estimation methods

Robust estimators based on medians, interquartile ranges, or trimmed criteria are commonly used. These methods reduce the influence of extreme observations and are often more stable than classical moment-based estimators. They are especially valuable when the data show Cauchy-like tail behavior.

7.2 Confidence intervals

Confidence intervals may be constructed using quantiles, likelihood methods, or resampling techniques. Since moments are unavailable, procedures based on asymptotic normality of the sample mean are inappropriate. Interval estimation for Cauchy parameters typically emphasizes robustness.

7.3 Hypothesis testing

Tests involving Cauchy data often use rank methods, likelihood ratios, or other nonparametric approaches. Classical tests that depend on sample means or variances may fail to perform well. The heavy-tailed structure requires careful choice of test statistic.

8 Applications

The Cauchy distribution appears in a range of applied settings, especially where extreme observations or angular constructions arise. It is also used as an idealized example in theoretical studies.

8.1 Physics and signal processing

In physics, the Cauchy distribution is closely related to resonance and line broadening, where it is also known as the Lorentz distribution. In signal processing, similar shapes can model spectral peaks with long tails. Its analytic simplicity makes it useful in theoretical treatments.

8.2 Random number generation

The tangent-of-uniform representation provides a straightforward way to generate Cauchy random variates. A uniform random angle can be transformed by the tangent function to produce a sample from the standard law. This method is simple, though the resulting values can be extremely large.

8.3 Modeling outliers and heavy tails

Cauchy models are sometimes used when data contain frequent outliers or exhibit unusually large deviations. In such settings, the distribution can serve as a stress test for statistical procedures. It is more often used as a benchmark than as a direct empirical model.

The Cauchy distribution is best understood by comparing it with other common laws and by noting its role in probability theory.

9.1 Comparison with normal distribution

Unlike the normal distribution, the Cauchy distribution has no finite mean or variance and much heavier tails. The normal law concentrates more strongly around its center, while the Cauchy allows larger deviations with far greater frequency. This contrast makes the Cauchy distribution a classic counterexample to intuition based on Gaussian behavior.

9.2 Comparison with Lorentz distribution

The Lorentz distribution is another name commonly used for the same density in physics. In many contexts, the two terms are interchangeable, although the name Lorentz is more frequent in spectral and resonance applications. Both refer to the same heavy-tailed shape.

9.3 Role in probability theory and statistics

The Cauchy distribution occupies a central place in probability theory because it illustrates the limitations of laws that depend on finite moments. It is also important in statistics as a case where robust methods outperform ordinary averaging. Its simplicity, symmetry, and pathological properties make it a standard teaching example.