The Gaussian error function (often denoted as erf) is a special mathematical function that appears in probability theory, statistics, and partial differential equations. It is defined as the integral of the standard normal distribution from zero to a given value, and is closely related to the cumulative distribution function of the normal distribution. The error function is used to describe the probability of a random variable falling within a certain range, and it also arises in heat conduction, diffusion processes, and the solution of certain differential equations. Its complementary counterpart, erfc, is defined as 1 − erf.
1.1 Integral definition of erf
The error function is defined by the integral
\[ \operatorname{erf}(x) = \frac{2}{\sqrt{\pi}} \int_{0}^{x} e^{-t^{2}} \, dt. \]
The factor \(\frac{2}{\sqrt{\pi}}\) normalizes the total area under \(e^{-t^{2}}\) over the real line to unity, so that \(\operatorname{erf}(\infty)=1\).
1.1.1 Relationship to the normal distribution
If \(\Phi\) denotes the cumulative distribution function (CDF) of the standard normal distribution (mean 0, variance 1), then
\[ \Phi(x) = \frac{1}{2}\left[1 + \operatorname{erf}\!\left(\frac{x}{\sqrt{2}}\right)\right]. \]
Conversely, \(\operatorname{erf}(x) = 2\,\Phi(x\sqrt{2}) - 1\).
1.1.2 Complementary error function (erfc)
The complementary error function is defined as
\[ \operatorname{erfc}(x) = 1 - \operatorname{erf}(x) = \frac{2}{\sqrt{\pi}} \int_{x}^{\infty} e^{-t^{2}} \, dt. \]
It satisfies \(\operatorname{erfc}(x) = 2\,\Phi(-x\sqrt{2})\) and is often used for tail probability calculations.
1.2 Series expansion and asymptotic behavior
1.2.1 Power series representation
For small \(x\), the error function can be expanded as
\[ \operatorname{erf}(x) = \frac{2}{\sqrt{\pi}} \sum_{n=0}^{\infty} \frac{(-1)^{n} x^{2n+1}}{n!\,(2n+1)}. \]
| This series converges for all finite \(x\) but is most useful for \( | x | < 1\). |
|---|
1.2.2 Asymptotic expansion for large arguments
For large positive \(x\), the complementary error function admits the asymptotic series
\[ \operatorname{erfc}(x) \sim \frac{e^{-x^{2}}}{x\sqrt{\pi}} \left(1 - \frac{1}{2x^{2}} + \frac{3}{4x^{4}} - \cdots \right), \]
which diverges but provides an excellent approximation when truncated at an appropriate term.
1.3 Symmetry and monotonicity
1.3.1 Odd function property
The error function is an odd function: \(\operatorname{erf}(-x) = -\operatorname{erf}(x)\). This follows directly from the definition and the evenness of the integrand.
1.3.2 Range and limits
\(\operatorname{erf}(x)\) is strictly increasing. Its range is \((-1,1)\): \(\lim_{x\to -\infty}\operatorname{erf}(x) = -1\), \(\lim_{x\to 0}\operatorname{erf}(x)=0\), \(\lim_{x\to +\infty}\operatorname{erf}(x)=1\).
2.1 Chebyshev and rational approximations
2.1.1 Minimax polynomial approximations
Many software libraries use minimax polynomial approximations on intervals such as \([0, \infty)\) to achieve near machine‑precision. For example, a rational function of the form
\[ \operatorname{erf}(x) \approx x \frac{a_0 + a_1 x^2 + a_2 x^4 + \cdots}{1 + b_1 x^2 + b_2 x^4 + \cdots} \]
can be optimized to minimize the maximum relative error.
2.1.2 Padé approximants
Padé approximants (ratios of polynomials) provide compact representations with good accuracy. A classic Padé form for \(\operatorname{erf}(x)\) or \(\operatorname{erfc}(x)\) is used in many numerical recipes, balancing speed and precision.
2.2 Implementation in computer algebra systems
2.2.1 Error function in libraries (e.g., math.erf)
Programming languages and environments (Python’s math.erf, MATLAB’s erf, SciPy’s scipy.special.erf) implement the error function via underlying C libraries that use minimax approximations or Chebyshev expansions.
2.2.2 High-precision algorithms
For arbitrary precision, algorithms based on the incomplete gamma function or continued fractions are employed. The asymptotic expansion may be used for large arguments, while the power series is applied for small arguments, with careful range reduction.
2.3 Table of selected values
| \(x\) | \(\operatorname{erf}(x)\) | |
|---|---|---|
| 0.0 | 0.000000 | |
| 0.5 | 0.520500 | |
| 1.0 | 0.842701 | |
| 1.5 | 0.966105 | |
| 2.0 | 0.995322 | |
| 3.0 | 0.999978 | |
| \(\infty\) | 1.000000 |
3.1 Relation to the cumulative distribution function (CDF) of the normal distribution
3.1.1 Conversion formula
As given in §1.1.1, the standard normal CDF \(\Phi\) and erf are related by
\[ \Phi(x) = \frac12\left[1 + \operatorname{erf}\!\left(\frac{x}{\sqrt{2}}\right)\right], \qquad \operatorname{erf}(x) = 2\Phi(x\sqrt{2}) - 1. \]
3.1.2 The probit function
The probit function is the quantile function of the standard normal: \(\Phi^{-1}(p) = \sqrt{2}\,\operatorname{erf}^{-1}(2p-1)\). Thus \(\operatorname{erf}^{-1}\) is directly related to the inverse CDF.
3.2 Relation to the imaginary error function (erfi)
The imaginary error function is defined as \(\operatorname{erfi}(x) = -i\, \operatorname{erf}(ix)\), or equivalently
\[ \operatorname{erfi}(x) = \frac{2}{\sqrt{\pi}} \int_{0}^{x} e^{t^{2}} \, dt. \]
It appears in expansions involving Dawson’s integral and in certain wave equations.
3.3 Relation to the incomplete gamma function
The error function can be expressed in terms of the lower incomplete gamma function:
\[ \operatorname{erf}(x) = \frac{1}{\sqrt{\pi}}\, \gamma\!\left(\frac12, x^{2}\right), \]
where \(\gamma(a,z) = \int_{0}^{z} t^{a-1} e^{-t} dt\). Similarly, \(\operatorname{erfc}(x) = \frac{1}{\sqrt{\pi}}\,\Gamma\!\left(\frac12, x^{2}\right)\) using the upper incomplete gamma function.
4.1 Gaussian integrals and tail probabilities
4.1.1 Confidence intervals
For a normally distributed variable, the probability that a value lies within \(k\) standard deviations of the mean is \(\operatorname{erf}(k/\sqrt{2})\). Common intervals: \(k=1\) gives about 68.3%, \(k=2\) about 95.5%, \(k=3\) about 99.7%.
4.1.2 Error bounds in signal processing
In digital communications, the bit error rate (BER) for Gaussian noise is often expressed via \(\operatorname{erfc}\). For example, the BER of a binary phase‑shift keying (BPSK) system is \(\frac12 \operatorname{erfc}(\sqrt{E_b/N_0})\).
4.2 Normal distribution and hypothesis testing
4.2.1 z‑scores and p‑values
| A z‑score \(z = (x-\mu)/\sigma\) converts to a p‑value via \(p = \frac12 \operatorname{erfc}( | z | /\sqrt{2})\). This is how standard normal tail probabilities are computed from erf. |
|---|
4.2.2 Quality control and six sigma
In the Six Sigma methodology, the defect rate corresponding to a process shift of \(\pm 1.5\sigma\) is computed using \(\operatorname{erfc}\). The target defect rate of 3.4 parts per million arises from a value of \(\operatorname{erfc}(4.5/\sqrt{2})\).
5.1 Heat equation and diffusion
5.1.1 Fundamental solution and error function
The solution of the one‑dimensional heat equation with an initial step function (Heaviside) involves the error function:
\[ u(x,t) = \frac12 \operatorname{erfc}\!\left(\frac{x}{2\sqrt{kt}}\right). \]
This describes the temperature profile after a sudden change at the boundary.
5.1.2 Stefan–Boltzmann problem
In moving‑boundary problems (e.g., melting of a solid), the interface position often follows the similarity variable \(\lambda = x / (2\sqrt{\alpha t})\), and the solution is expressed via the error function.
5.2 Error function in optics and signal propagation
5.2.1 Fresnel integrals and diffraction
The Fresnel integrals \(C(x)\) and \(S(x)\) are related to the error function of complex argument. Specifically, \(C(x) + i S(x) = \frac{1+i}{2} \operatorname{erf}\!\left(\frac{\sqrt{\pi}(1-i)x}{2}\right)\).
5.2.2 Communication theory (bit error rate)
As noted in §4.1.2, \(\operatorname{erfc}\) is fundamental in calculating the probability of symbol error in additive white Gaussian noise (AWGN) channels. For example, the symbol error rate for QAM or PSK is directly proportional to \(\operatorname{erfc}\).
6.1 Generalized error functions
6.1.1 Multiple‑argument error functions
Generalizations such as \(\operatorname{erf}(x_1, x_2, \dots)\) appear in multivariate normal distribution integrals. For instance, the bivariate normal integral can be expressed via an integral of an error function.
6.1.2 Incomplete error functions
The incomplete error functions \(\operatorname{erf}(z, a)\) and \(\operatorname{erfc}(z, a)\) are defined with lower or upper limits different from zero/infinity, extending the standard definitions.
6.2 Derivatives and integrals involving erf
6.2.1 Derivative formula
\[ \frac{d}{dx} \operatorname{erf}(x) = \frac{2}{\sqrt{\pi}} e^{-x^{2}}. \]
Higher derivatives can be expressed in terms of Hermite polynomials.
6.2.2 Antiderivatives and repeated integrals
The indefinite integral \(\int \operatorname{erf}(x) dx = x\,\operatorname{erf}(x) + \frac{e^{-x^{2}}}{\sqrt{\pi}} + C\). Repeated integrals of the error function (iterated error functions) appear in some series expansions.
6.3 Complex error function (Faddeeva function)
The Faddeeva function \(w(z) = e^{-z^{2}}\operatorname{erfc}(-iz)\) is widely used in spectroscopy and plasma physics. It satisfies \(w(z) = e^{-z^{2}}(1 + \frac{2i}{\sqrt{\pi}} \int_{0}^{z} e^{t^{2}} dt)\) and is the basis for the complex error function.
7.1 Origins in probability (De Moivre, Laplace, Gauss)
The integral \(\int e^{-x^{2}} dx\) was studied by Abraham de Moivre in the early 18th century in the context of approximating binomial probabilities. Pierre‑Simon Laplace and Carl Friedrich Gauss later developed the normal distribution. Gauss explicitly used the error function in his work on least squares (1809).
7.2 Adoption in mathematical physics
The error function appeared in the solution of the heat equation by Joseph Fourier (1822) and in the theory of errors. Over the 19th century it became standard in diffusion theory and the theory of elasticity.
7.3 Standardization of notation (erf, erfc)
The notations \(\operatorname{erf}\) and \(\operatorname{erfc}\) were introduced by J. W. L. Glaisher in 1871. They were popularized by tables and textbooks of the early 20th century, and became standard in special function handbooks such as Abramowitz and Stegun (1964).