1 Definition and basic properties
1.1 Relation to the gamma function
The digamma function, denoted \( \psi(x) \), is defined as the logarithmic derivative of the gamma function \( \Gamma(x) \): \[ \psi(x)=\frac{d}{dx}\ln\Gamma(x)=\frac{\Gamma'(x)}{\Gamma(x)}. \] This definition makes \( \psi \) the natural tool for converting multiplicative structure in \( \Gamma \) into additive (derivative-based) behavior. Many manipulations involving \( \Gamma \) products become simpler after taking logarithms and differentiating, producing expressions in \( \psi \) or related higher derivatives.
1.2 Domain, poles, and analytic structure
Because \( \Gamma(x) \) has simple poles at the non-positive integers, the ratio \( \Gamma'(x)/\Gamma(x) \) inherits corresponding singularities. The digamma function is meromorphic on \( \mathbb{C} \) with simple poles at \[ x = 0,-1,-2,\ldots \] and is analytic elsewhere. Near each pole, the growth of \( \psi(x) \) is governed by the residue and principal part determined by the local behavior of \( \Gamma(x) \).
1.3 Recurrence relation (shift formula)
A fundamental identity shifts the argument by one: \[ \psi(x+1)=\psi(x)+\frac{1}{x}. \] This follows from the gamma recurrence \( \Gamma(x+1)=x\Gamma(x) \). The shift formula is repeatedly used to reduce evaluations to a convenient region, to derive special values, and to express differences such as \( \psi(b)-\psi(a) \) in terms of rational sums.
1.4 Special values at integers and half-integers
For positive integers \(n\), the digamma function connects to harmonic numbers: \[ \psi(n)=H_{n-1}-\gamma, \] where \(H_{n-1}=\sum_{k=1}^{n-1}\frac{1}{k}\) and \( \gamma \) is the Euler–Mascheroni constant. For half-integers, closed forms involve \( \gamma \), logarithms, and harmonic-number-like sums with alternating terms; for example, \[ \psi\!\left(\tfrac12\right)=-\gamma-2\ln 2, \] and the recurrence then extends these values to \( \psi(n+\tfrac12) \).
2 Representations
2.1 Integral representations
A commonly used integral form for \( \Re(x)>0 \) is \[ \psi(x)=-\gamma+\int_{0}^{\infty}\left(\frac{e^{-t}}{t}-\frac{e^{-xt}}{1-e^{-t}}\right)\,dt. \] Such formulas provide analytic continuation by meromorphic continuation in \(x\) and are also practical for proving identities, establishing monotonicity/convexity properties, and deriving expansions.
Another standard representation expresses \( \psi \) via a Laplace-type integral: \[ \psi(x)=\ln x - \int_{0}^{\infty}\frac{1}{(x+t)}\,\left(\frac{1}{e^{t}}- \text{(normalization)}\right)\,dt, \] with the exact normalization chosen to ensure convergence; these variants emphasize \( \ln x \) behavior for large \(x\).
2.2 Series expansions
2.2.1 Hurwitz zeta and related expansions
The digamma function is expressible in terms of the Hurwitz zeta function \( \zeta(s,a) \). One useful relation connects \( \psi \) to the derivative of \( \zeta \) at \(s=0\): \[ \psi(a) = \frac{d}{da}\ln \Gamma(a) = -\gamma + \sum_{n=0}^{\infty}\left(\frac{1}{n+1}-\frac{1}{n+a}\right), \] and more generally representations in terms of \( \zeta(s,a) \) allow analytic continuation and controlled expansions by leveraging known properties of the zeta function. In practice, these zeta-linked series are especially convenient when \(a\) is not too close to a pole and when convergence acceleration is applied.
2.2.2 Expansion near poles and regularization viewpoints
Because \( \psi \) has simple poles at non-positive integers, local expansions around \(x=-m\) (\(m\in\mathbb{Z}_{\ge 0}\)) separate the principal part from the regular remainder. One writes \[ \psi(x)= -\frac{1}{x+m} + \text{(finite part depending on }m\text{)} + O(x+m). \] This “finite part” can be expressed using harmonic-number data and logarithmic terms via analytic continuation of \( \Gamma \). These expansions are useful for regularization schemes in analysis and for symbolic simplification, where isolating singularities prevents loss of precision.
2.3 Asymptotic expansion
2.3.1 Stirling-type expansions for large arguments
| For large \( | x | \) with \( | \arg x | <\pi \), \( \psi(x) \) admits an asymptotic series derived from Stirling’s expansion of \( \Gamma(x) \). A classical form is |
|---|
\[ \psi(x)\sim \ln x - \frac{1}{2x} - \sum_{k=1}^{\infty}\frac{B_{2k}}{2k\,x^{2k}}, \] where \(B_{2k}\) are Bernoulli numbers. This shows that \( \psi(x) \) approaches \( \ln x \) with a systematic correction hierarchy in inverse powers of \(x\).
2.3.2 Error bounds and practical truncation
In numerical work, one truncates the asymptotic series after a finite number of terms. Although asymptotic series do not converge in general, truncation at an appropriate point yields small error. Error estimates are typically based on the next neglected term and bounds derived from alternating behavior of the Bernoulli-number sequence for certain argument ranges. Practical algorithms choose the number of terms using magnitude and argument checks to balance speed against accuracy.
3 Functional identities
3.1 Reflection formula
The digamma function satisfies a reflection identity that relates values at \(x\) and \(1-x\): \[ \psi(1-x)-\psi(x)=\pi \cot(\pi x). \] This formula is central for converting evaluations across complementary arguments and for understanding symmetry around \(x=\tfrac12\). It also helps reveal pole structure by matching the cotangent’s singularities.
3.2 Multiplication theorems
Gauss-type multiplication theorems generalize the shift relation by scaling the argument. For an integer \(m\ge 1\), \[ \psi(mx)=\ln m+\frac{1}{m}\sum_{k=0}^{m-1}\psi\!\left(x+\frac{k}{m}\right). \] These identities allow reduction of complex expressions into sums of digamma values with simpler arguments, and they are useful for simplifying series that involve rational shifts.
3.3 Symmetry and transformations
Beyond the reflection and multiplication formulas, digamma transformations frequently use combinations of shifts to move arguments into a preferred domain. Typical manipulations use the recurrence relation to express \[ \psi(x+n)=\psi(x)+\sum_{k=0}^{n-1}\frac{1}{x+k}, \] or to rewrite differences into harmonic-type sums. Such transformations are algebraically stable when the chosen direction avoids crossing poles.
3.4 Connection to polygamma functions
The polygamma functions \( \psi^{(n)}(x) \) are derivatives of \( \psi \): \[ \psi^{(n)}(x)=\frac{d^{n+1}}{dx^{n+1}}\ln\Gamma(x). \] In particular, \( \psi^{(1)}(x) \) is the trigamma function, which can be represented by series of positive terms for \(x>0\). Many identities for \( \psi \) extend to polygamma functions by differentiating the functional relations where applicable, yielding families of transformation rules for higher derivatives.
4 Connections to other mathematical objects
4.1 Harmonic numbers and generalized harmonic numbers
For integers, harmonic numbers provide direct special values of the digamma function. The relation \[ \psi(n)=H_{n-1}-\gamma \] connects analytic special functions with finite sums. More generally, the appearance of \( \psi \) in differences such as \( \psi(n+a)-\psi(a) \) yields generalized harmonic sums involving reciprocals of arithmetic progressions, often expressible through rational combinations of zeta-like quantities for specific parameters.
4.2 Logarithmic derivatives in combinatorics
Logarithmic derivatives frequently appear when differentiating factorial-like expressions in generating functions. Since \( \Gamma \) extends factorials, ratios of \( \Gamma \) functions and derivatives lead naturally to digamma terms. In combinatorics, this occurs when analyzing asymptotics of binomial coefficients, Stirling-approximation corrections, and normalization constants in occupancy or partition models.
4.3 Links to the Hurwitz zeta function
Many zeta-connected expansions can be organized by expressing \( \psi \) through \( \zeta(s,a) \) and its derivatives. This link is particularly powerful because \( \zeta \) enjoys functional equations and well-developed analytic continuation theory. As a result, digamma expressions often inherit structured expansions in powers of \(a\) or in large-parameter limits.
4.4 Derivatives of the gamma and beta functions
4.1 Harmonic numbers and generalized harmonic numbers
The beta function \(B(x,y)=\frac{\Gamma(x)\Gamma(y)}{\Gamma(x+y)}\) has derivatives that involve \( \psi \). Differentiating logarithmically, \[ \frac{\partial}{\partial x}\ln B(x,y)=\psi(x)-\psi(x+y), \] and similarly for \(y\). Hence derivatives of \(B\) translate into differences of digamma values, a pattern that appears throughout probability (via gamma and beta normalizations) and in integral evaluations.
5 Computation and numerical aspects
5.1 Stable evaluation strategies
Direct evaluation from definitions is typically avoided. Instead, implementations use combinations of recurrence, reflection, and asymptotic series. A common strategy is to shift \(x\) into a region where series converge rapidly, then evaluate using either:
- a power series / zeta-based series for moderate \(x\), or
- an asymptotic expansion for large \(x\),
while carefully handling neighborhoods of poles.
5.2 Recurrence acceleration and truncation choices
Because \[ \psi(x+1)=\psi(x)+\frac{1}{x}, \]
| one can move \(x\) to a target interval by repeatedly applying the recurrence. This reduces sensitivity and improves convergence of subsequent expansions. Truncation choices for asymptotic series depend on \( | x | \) and on desired precision; selecting a term count near the point where the terms stop decreasing in magnitude often yields good practical accuracy. |
|---|
5.3 Approximations for small and large arguments
For large arguments, \( \psi(x)\approx \ln x - \frac{1}{2x} \) is a quick first estimate, improved by adding Bernoulli corrections. For small arguments away from poles, one uses either recurrence to shift upward to a safer region or integral/series representations that remain well-behaved. When \(x\) is near a pole, algorithms isolate the singular component \( -1/(x+m) \) and compute the regular remainder with stable formulas.
5.4 Software/library conventions and edge cases
Numerical libraries typically:
- return complex values consistently with the chosen branch of the logarithm,
- treat poles as floating-point overflows or as signed infinities depending on the platform,
- implement argument reduction using reflection when it improves accuracy.
Edge cases include purely real inputs near negative integers, where cancellation may occur if singular parts are not separated, and complex inputs near the branch cut of \(\arg x\).
6 Applications and typical use cases
6.1 Derivatives in special-function calculus
Because \( \psi \) is the logarithmic derivative of \( \Gamma \), it naturally appears in differentiation of expressions containing \( \Gamma \) factors. This is common in symbolic manipulation systems, in analytic derivations of parameter derivatives, and in calculations of sensitivity of normalization constants in integral transforms.
6.2 Summations involving rational and harmonic terms
Digamma identities convert sums of reciprocals and rational functions into compact forms. For example, finite sums over arithmetic progressions can be expressed using differences of digamma values. This reduces computational complexity and improves clarity in analytic proofs, particularly when sums are long or when parameters vary continuously.
6.3 Statistical distributions and expectations (non-controversial overview)
In probability, gamma and beta distributions use \( \Gamma \) functions for normalization. Expectations of logarithms of random variables often reduce to digamma evaluations; for instance, if a variable follows a gamma-family law with shape parameter, the expected log of the variable typically equals a digamma term minus a logarithm of a scale. Similar relationships occur in beta-family settings via beta-function derivatives, making \( \psi \) a standard component in likelihood derivatives and variational update rules.
6.4 Integral transforms where ψ(x) arises
Integral transforms that involve gamma and beta kernels frequently produce digamma terms after differentiation with respect to parameters. In Mellin-transform contexts, where gamma functions appear after transforming powers, parameter derivatives yield \( \psi \) and polygamma functions. This is especially relevant in evaluating moments, asymptotic coefficients, and parameterized integrals where logarithmic factors are present.