# Dickman function

The function defined on by the initial condition for and by the differential-delay equation for . Interest attaches to this function because of its connection to "smooth" numbers, i.e. numbers that are the product of many small prime numbers. Let denote the number of positive integers less than or equal to and free of prime divisors greater than . When is much larger than , it is a simple matter of inclusion-and-exclusion counting (cf. also Inclusion-exclusion formula) to show that , where denotes a prime number. But the error terms grow rapidly, and the "main" term gives the wrong answer in the ranges of greatest interest, including the case when is comparable to a fixed fractional power of . For this case, K. Dickman found that . If, in place of the restriction for one takes the condition , the resulting is approximated by , where is the Buchstab function, defined by , , and , , where for the right-hand derivative has to be taken, [a1]. Unlike , oscillates and tends to a positive limit, equal to .

There are two combinatorial identities linking the Dickman function to . Early work is based on the Buchstab identity: With denoting a prime number, for , The usual heuristic device of replacing a sum over prime numbers by an integral with "prime density" and replacing with leads to an identity which, when and , simplifies to an integral equivalent to the definition of . N.G. de Bruijn carried this idea to its limits in [a2], but accuracy suffers when large and comparable estimated quantities must be subtracted. The more recent Hildebrand identity involves only additions and has the further advantage that the second input is the same throughout: Applications require estimates uniform in ; the best known estimate along these lines, due to A. Hildebrand and based on the identity above, is uniformly in and . There are similar results for algebraic integers, [a3].

There are also results concerning the number of smooth integers in an interval, and concerning the distribution of smooth integers into congruence classes [a4], [a5].

The Riemann hypothesis (cf. Riemann hypotheses) implies [a8].

The analytical properties of are reasonably well understood; calculus, analysis of the Laplace transform, and the saddle-point method are the key tools. The first extensive analysis is due to De Bruijn, and the function is sometimes termed the Dickman–De Bruijn function. One has [a10]:

a) for (so that is positive for all positive , and hence, from the definition, decreasing);

b) is log-concave, that is, for and (K. Alladi);

c) The Laplace transform  d) as .

The Dickman function is one of a parameterized family of related functions , [a12], and a wider class of similar delay-differential equations has been studied in [a7]. A quick and simple bit of Mathematica code suffices to calculate to reasonable accuracy in the interval using step-size . This code calculates a table of values of at intervals of length , working back recursively into :

<tbody> </tbody>  , ; ;  