For real and , the lower and upper incomplete gamma functions are respectivelyThey divide the Gamma function integral at , so . The normalized ratios and sum to1. Lowercase and uppercase distinguish the two tails; using one symbol for both would reverse which part is omitted.[ref]
For example, and . Thus is the cumulative probability and the tail probability for a gamma distribution of unit scale. Integration by parts gives , a recurrence useful for both numerical evaluation and endpoint asymptotic expansions.
For , this integral is an entire function of : the exponential series converges uniformly on bounded subsets after multiplication by the integrable bound . ThusFor positive real , a change of variables gives , where here denotes the lower incomplete gamma function. The singular-looking factor is removed by the integral over . Applying integration by parts to the derivative of gives ; differentiating and eliminating proves .
For fixed real and , integration by parts in the upper incomplete gamma function givesRepeating times yields the exact identityFor fixed as , the last term is , giving an asymptotic expansion in inverse endpoint powers.[ref]
This is an exact remainder formula before it becomes an asymptotic statement. If is a positive integer, a coefficient eventually vanishes and the expansion terminates. For instance . For noninteger , the expansion need not converge as an infinite series; its purpose is controlled finite truncation.
For fixed and real , it is the upper incomplete gamma function that has the endpoint-tail expansionThe empty product is one. Repeated integration by parts gives each coefficient and an integral remainder. For a positive integer , the product eventually vanishes and the expansion terminates. The lower function instead satisfies for ; confusing the two functions would incorrectly make the lower function decay to zero.[ref]
For a fixed real and a real shape parameter , the upper incomplete gamma function satisfiesThe first equivalence says that a fixed lower cutoff eventually removes a negligible fraction of the whole Gamma integral; the second is Stirling's formula. For ,Dividing by the Stirling scale makes this bound tend to zero. The case is exact. This proof also explains why a cutoff growing proportionally to requires a different approximation: it can enter the region where the mass of the integrand is concentrated. The symbol in the conclusion is uppercase for the upper tail, rather than lowercase .
For a fixed and real , the upper incomplete gamma function has leading endpoint approximationSet in its defining integral:The phase is increasing for , with derivative at the endpoint. The leading endpoint form of Laplace's method gives amplitude divided by , proving the formula.[ref]
Higher terms come from expanding the transformed amplitude near that endpoint. This expansion is not uniform as , where the endpoint approaches the stationary point of the phase. A transition approximation is then needed rather than substituting into a denominator that vanishes.
Articles by others on the same topic
The incomplete gamma function is a mathematical function that generalizes the gamma function, which itself is a fundamental function in mathematics, particularly in the fields of statistics and probability theory. The incomplete gamma function is useful in various applications, including statistical distributions and hypothesis testing. The incomplete gamma function is defined in two forms: the lower incomplete gamma function and the upper incomplete gamma function.