aa r X i v : . [ m a t h . N T ] J u l ABSTRACT FACTORIALS
ANGELO B. MINGARELLI
Abstract.
A commutative semigroup of abstract factorials is defined in thecontext of the ring of integers. We study such factorials for their own sake,whether they are or are not connected to sets of integers. Given a subset X ⊆ Z + we construct a “factorial set” with which one may define a multitudeof abstract factorials on X . We study the possible equality of consecutivefactorials, a dichotomy involving the limit superior of the ratios of consecutivefactorials and we provide many examples outlining the applications of theensuing theory; examples dealing with prime numbers, Fibonacci numbers,and highly composite numbers among other sets of integers. One of our resultsstates that given any abstract factorial the series of reciprocals of its factorialsalways converges to an irrational number. Thus, for example, for any positiveinteger k the series of the reciprocals of the k -th powers of the cumulativeproduct of the divisors of the numbers from 1 to n is irrational. Introduction
The study of generalized factorials has been a subject of interest during the pastcentury culminating in the appearance of many treatises that fostered widespreadapplications. An introduction and review of this area has already been given inthe recent paper [3] and so we shall not delve into the historical matter any furtherunless our applications require it so, referring the reader to [3] for more information.In this paper we restrict ourselves to the ring of integers only. Abstract factorialsare defined as maps ! a : N → Z + satisfying very few conditions, conditions that areverified by apparently all existing notions of a generalized factorial in this context.As a consequence of the results herein we obtain, among other such results, theirrationality of the following numbers and classes of numbers, where b, q, k ∈ Z + are arbitrary, ∞ X n =1 Q ni =1 p ik ⌊ n/i ⌋ , ∞ X n =1 n ! q P nj =1 d ( j ) , ∞ X n =1 n ! q P nj =1 σ k ( j ) , Date : July 10, 2012 - Dedicated to the memory of my teacher, Professor Hans Heilbronn.2010
Mathematics Subject Classification.
Key words and phrases.
Abstract factorial, factorial, factorial sets, factorial sequence, irra-tional, divisor function, Von Mangoldt function, cumulative product, Hardy-Littlewood conjec-ture, prime numbers, highly composite numbers, Fibonacci numbers.This research is partially supported by an NSERC Canada Discovery Grant. and ∞ X n =1 b nk ( bn )! k , ∞ X n =1 n ! F ( n ) k , ∞ X n =1 q P nj =1 d ( j ) α ( n ) k ∞ X n =1 p n ! , where α ( n ) = Q ni =1 i ⌊ n/i ⌋ is the cumulative product of all the divisors from 1 to n , F ( n ) is the product of the first n Fibonacci numbers, and p n is the n -th prime.Furthermore, there is a sequence of highly composite numbers h n such that ∞ X n =1 Q ni =1 h i ⌊ n/i ⌋ is irrational. In addition, if f : Z + × Z + → Z + satisfies a concavity condition in itsfirst variable, i.e., for any x, y, q , we have f ( x + y, q ) ≥ f ( x, q ) + f ( y, q ), then, forany q ∈ Z + , ∞ X n =1 q f ( n,q ) n !is also irrational. We also show that we may choose f ( n, q ) = (cid:0) n + q − q (cid:1) in theprevious result.An abstract factorial will be denoted simply by the notation n ! a , the usual factorialfunction being denoted by n !. Other unspecified abstract factorials will be indexednumerically (e.g., n ! , n ! , . . . .)We always assume that X is a non-empty set of non-zero integers. For the sake ofsimplicity assume that X ⊆ Z + , although this is, strictly speaking, not necessaryas the constructions will show. Using the elements of X we construct a new set(generally not unique) dubbed a factorial set of X . We will see that any factorialset of X may be used out to construct infinitely many abstract factorials (seeSection 3). This construction of generalized (abstract) factorials of X should beseen as complementary to that of Bhargava, [3]. Furthermore, there is enoughstructure in the definition of these abstract factorials that, as a collective, theyform a semigroup under ordinary multiplication.By their very nature abstract factorials should go hand-in-hand with binomial co-efficients (Definition 1). For example, Knuth and Wilf [13] define a generalizedbinomial coefficient by first starting with a positive integer sequence C = { C n } andthen defining the binomial coefficient as a “falling” chain type product (cid:18) n + mm (cid:19) C = C m + n C m + n − . . . C m +1 /C n C n − . . . C . In this case, the quantity n ! a = n ! C C · · · C n always defines an abstract factorialaccording to our definition.In Section 2 we give the definition of an abstract factorial (Definition 1), give theirrepresentation (Proposition 4) and show that, generally, consecutive equal factorialsmay occur. In fact, strings of three or more consecutive equal factorials cannot occur(Lemma 8). Of special interest is the quantity defined by the ratio of consecutivefactorials (2) for which there exists a dichotomy, i.e., there always holds either (3)or (4) (Lemma 10). Cases of equality in both (3) or (4) are exhibited by specific BSTRACT FACTORIALS 3 examples (Proposition 12 in the former case, and use of Bhargava’s factorials forthe set of primes [3] in the latter case).Generally, given a set X we find its factorial sets (Section 3). We then show thatfor any abstract factorial (whether or not it should arise from a set) the sum ofthe reciprocals of its generalized factorials is always irrational (Section 4). Anapplication of the semigroup property (Proposition 3) and the global irrationalityresult (from Lemma 24 and Lemma 27) implies that if ! , ! ,. . . ,! k is any collectionof abstract factorials, s i ∈ N , i = 1 , , . . . , k , not all equal to zero, then ∞ X n =0 Q kj =1 n ! s j j is irrational (Theorem 28). As a consequence of our theory we also obtain the irra-tionality of the series of reciprocals of the generalized factorials (and their powers)for the set of primes in [3], (see Corollary 26) and the other series displayed earlier.In Section 5 we consider an inverse problem that may be stated thus: Given anyabstract factorial n ! a , does there exist a set X such that the sequence of generalizedfactorials { n ! a } ∞ n =0 coincides with one of the factorial sets of X ? If there is sucha set X , it will be called a primitive of the abstract factorial in question. It isnoteworthy that such primitives are usually not unique.In this direction we find that a primitive of the ordinary factorial function, n !, isgiven simply by the exponential of the Von Mangoldt function i.e., X = { e Λ( m ) : m = 1 , , . . . } . In other words, the ordered set X = { , , , , , , , , , , , , , . . . } whose n -th term is given by b n = e Λ( n ) has a factorial set whose elements coincidewith the sequence of factorials of the ordinary factorial function (Theorem 36). Wefind (Theorem 37) that Bhargava’s generalized factorial for the set of primes alsohas a primitive X = { , , , , , , , , , , , , . . . } where every term here is the product of at most two primes. Still, it has a factorialfunction that agrees with the generalized factorial for the set of primes in [3]. Thus,generally speaking, there are a number of ways in which one may associate a setwith an abstract factorial and conversely.In Section 6 we give some applications of the foregoing theory. We also introducethe notion of a self-factorial set, that is, basically a set whose elements are eitherthe factorials of some abstract factorial, or are so when multiplied by n !. We findsome abstract factorials of sets such as the positive integers, X = Z + , and showthat one of its factorial sets is given by the set { n ! a : n = 0 , , , . . . } where (seeExample 46) n ! a = n Y i =1 i i ⌊ n/i ⌋ . Furthermore, in Example 42 we show that one of the factorial sets of the set { , q, q, q, . . . } where q ∈ Z + , q ≥
2, is given by the set { B n } where B n = q P nk =1 d ( k ) ANGELO B. MINGARELLI where d ( n ) is the usual divisor function.Combining the preceding with the results of Section 4 we also obtain that the seriesof reciprocals of the k -th powers ( k ≥
1) of the cumulative product of all the divisorsof the integers from 1 to n , i.e., ∞ X n =1 / n Y i =1 i k ⌊ n/i ⌋ , is irrational (see Example 46 and Remark 47).In the same spirit we show in Example 48 that for q ∈ Z + , q ≥
2, the set { q n : n ∈ N } has a factorial set { B n } where B n = q P nk =1 σ ( k ) where σ ( n ) is the sum of the divisors of n , a result that can be extended to the caseof sets of integers of the form { q n k } for given k ≥ X = q Z + , q >
0. Here, one of the factorial sets of X is givenby numbers of the form B n = q P nk =1 d ( k ) n Y i =1 i ⌊ n/i ⌋ , where the product on the right is once again the cumulative product arithmeticfunction defined above (see Remark 47).Subsections 6.1-6.2 are devoted to questions involving prime numbers in our set X , their (abstract) factorials, and the problem of determining whether a functionarising from the ordinary factorial of the n-th prime number is, indeed, an abstractfactorial. This latter question is, in fact, related to an unsolved problem of Hardyand Littlewood dealing with the convexity of the prime-counting function π ( x ).We show that the Hardy-Littlewood conjecture on the prime counting function, π ( x ), i.e., that for all x, y there holds π ( x + y ) ≤ π ( x ) + π ( y ) , implies that p n ≥ p k + p n − k − , ≤ k ≤ n − , where p n is the n -th prime, and that this inequality in turn implies that the “primefactorial function” f : N → Z + defined by f (0) = 1, f (1) = 1 and f ( n ) = p n − !, n ≥
2, is an abstract factorial. Although said conjecture may be false according tosome, it may be the case that the above inequality holds.We recall the definition of a highly composite number (hcn): A number n is said tobe highly composite if d ( m ) < d ( n ) whenever m < n , where d is the usual divisorfunction. After proceeding to the calculation of a factorial set of the set of primes,we note that the first six numbers of this set are actually highly composite numbersand, in fact, we prove that these are the only ones (Proposition 54). BSTRACT FACTORIALS 5
An application of the theory developed here allows us to derive that for everypositive integer k , the series ∞ X n =0 / ( p ⌊ n/ ⌋ p ⌊ n/ ⌋ p ⌊ n/ ⌋ · · · p ⌊ n/n ⌋ n ) k is irrational.In Subsection 6.3 we show, in particular, that given any positive integer m there isa highly composite number (hcn), N , such that m ! | N . We then find factorial setsof the set of hcn and show that they are all self-factorial. Using this it is shownthat there exists a sequence { h n } of hcn such that for any k ∈ Z + , ∞ X n =1 Q ni =1 h ik ⌊ n/i ⌋ is irrational (Proposition 59). This is one of the few results dealing with the irra-tionality of series involving hcn.We end the paper with a number of remarks. A brief note on the representationsof abstract factorials in terms of possible solutions of the Stieltjes moment problem(Section 7) is given at the end. The idea here is to find integral representations ofthese abstract factorials akin to the usual representation of the ordinary factorial interms of the Gamma function. Initial results in this direction indicate that variousclasses of abstract factorials admit unique integral representations as solutions of amoment problem. We note that not all abstract factorials admit such representa-tions. We also produce a few more simple irrationality criteria based on the resultsherein, such as the ones given at the outset, and give a simple proof that the sumof the reciprocals of the factorials of the primes is irrational.2. Preliminaries
In the sequel the symbols
X, I will always stand for non-empty subsets of Z , notcontaining 0, either may be finite or infinite, whose elements are not necessarily dis-tinct (e.g., thus the set X = { , q, q, q, . . . } is considered an infinite set). When thecontext is clear we will occasionally use the words sequence and sets interchange-ably. Definition 1.
An abstract (or generalized) factorial is a function ! a : N → Z + thatsatisfies the following conditions: (1) 0! a = 1 , (2) For every non-negative integers n, k , ≤ k ≤ n the generalized binomialcoefficients (cid:18) nk (cid:19) a := n ! a k ! a ( n − k )! a ∈ Z + , (3) For every positive integer n , n ! divides n ! a . Remark 2.
Since, by hypothesis (2) above, (cid:0) n +1 n (cid:1) a ∈ Z + for every n ∈ N thesequence of abstract factorials n ! a is non-decreasing. ANGELO B. MINGARELLI
Another simple consequence of the definition is,
Proposition 3.
The collection of all abstract factorials forms a commutative semi-group under ordinary multiplication.
Terminology:
In the sequel an abstract factorial function will be called simply a factorial function or an abstract factorial and its values will be referred to simplyas its factorials (or generalized factorials for emphasis), unless otherwise specified.Of course the ordinary factorial function n ! is an abstract factorial as is the functiondefined by setting n ! a := 2 n ( n +1) / n !. The factorial function defined in [3], forarbitrary sets X is also a factorial function (see Example 15). In addition, if C = { C n } is a positive integer sequence and we assume as in [13] that the binomialcoefficient (cid:18) n + mm (cid:19) C = C m + n C m + n − . . . C m +1 C n C n − . . . C is a positive integer for every n, m ∈ N , then there is an associated abstract factorial! a with these as binomial coefficients that is, the one defined by setting 0! a = 1 and n ! a = C C · · · C n provided n ! | C C · · · C n for every n ∈ Z + . On the other hand, if n ! does not divide C C · · · C n for every n we can still define another abstract factorial by writing n ! a = n ! C C · · · C n . Its binomial coefficients are now of the form (cid:18) n + mm (cid:19) a = (cid:18) n + mm (cid:19) C (cid:18) n + mm (cid:19) where the last binomial coefficient is the usual one. These generalized or abstractbinomial coefficients are necessarily integers because of the tacit assumption madein [13] on the binomial coefficients appearing in the middle of the previous display.All of our results below apply in particular to either one of the two precedingfactorial functions.Another consequence of Definition 1 is the following, Proposition 4.
Let ! a be an abstract factorial. Then there is a positive integersequence h n with h = 1 and such that for each n ∈ N , h k h n − k (cid:12)(cid:12)(cid:12)(cid:12) h n (cid:18) nk (cid:19) , k = 0 , , . . . , n. (1) Conversely, if there is a sequence of positive integers h n satisfying (1) and h = 1 ,then the function ! a : N → Z + defined by n ! a = n ! h n is an abstract factorial. Corollary 5.
Let h n ∈ Z + be such that h = 1 , h k h n − k | h n , for all k = 0 , , . . . , n, and for every n ∈ N . Then n ! a = n ! h n is an abstract factorial. In Section 3 below we consider those abstract factorials induced by those sequences h n such that h k h n − k | h n , for all k = 0 , , . . . , n, and for every n ∈ Z + . Suchsequences form the basis for the notion of a “factorial set” as we see below. BSTRACT FACTORIALS 7
Observe that h n is a constant sequence satisfying (1) if and only if h n = 1 for all n , that is, if and only if the abstract factorial reduces to the ordinary factorial.Note that distinct abstract factorials functions may have the same set of binomialcoefficients ; for example, if b ∈ Z + and n ! a = n ! b n , for every n , then the binomialcoefficients of this factorial function and the usual factorial function are identical.The reason for this lies in the easily verifiable identity n ! a = 1! n a n Y m =1 (cid:18) mm − (cid:19) a , valid for any abstract factorial. Thus, it is the value of 1! a that determines whetheror not an abstract factorial is determined uniquely by a knowledge of its binomialcoefficients.One of the curiosities of abstract factorials lies in the possible existence of equalconsecutive factorials . Definition 6.
Let ! a be an abstract factorial. By a pair of equal consecutivefactorials we mean a pair of consecutive factorials such that, for some k ≥ , k ! a = ( k + 1)! a . Remark 7.
Definition 6 is not vacuous as we do not tacitly assume that thefactorials form a strictly increasing sequence (cf., Example 16 and Proposition 12below). In addition, given an abstract factorial it is impossible for 1! a = 2! a . (Thisis because (cid:0) (cid:1) a must be an integer, which of course can never occur since 2! a mustbe at least 2.)Such equal consecutive factorials, when they exist, are connected to the propertiesof ratios of nearby factorials. We adopt the following notation for ease of exposition:For a given integer k and for a given factorial function ! a , we write r k = ( k + 1)! a k ! a . (2)Since generalized binomial coefficients are integers by Definition 1, r k is an integerfor every k = 0 , , , . . . . The next result shows that strings of three or more equalconsecutive factorials cannot occur. Lemma 8.
There is no abstract factorial with three consecutive equal factorials.
Knowing that consecutive equal factorials must occur in pairs if they exist at allwe get,
Lemma 9.
Given an abstract factorial ! a , let a = 2 . If r k = 1 for some k ≥ ,then r k − ≥ . The next result gives a limit to the asymptotics of sequences of ratios of consecutivefactorials defined by the reciprocals of the r k . These ratios do not necessarily tendto zero as one may expect (as in the case of the ordinary factorial), but may havesubsequences approaching non-zero limits! ANGELO B. MINGARELLI
Lemma 10.
For any abstract factorial, either lim sup k →∞ r k = 1 , (3) or lim sup k →∞ r k ≤ / , (4) the upper bound in (4) being sharp, equality being attained in the case of Bhargava’sfactorial for the set of primes (see the proof of Corollary 26). Definition 11.
An abstract factorial whose factorials satisfy (3) will be called exceptional.Note:
Using the generalized binomial coefficients (cid:0) n +1 n (cid:1) a it is easy to see that anecessary condition for the existence of such exceptional factorial functions is that1! a = 1. The question of their existence comes next. Proposition 12.
The function ! a : N → Z + defined by a = 1 , a = 1 andinductively by setting ( n + 1)! a = n ! a whenever n is of the form n = 3 m − forsome m ≥ , and n ! a = ( n ! ( n + 1)! Q n − j =1 ( n − j )! a , if n is of the form n = 3 m − ,n ! Q n − j =1 ( n − j )! a , if n is of the form n = 3 m + 1 . is an exceptional factorial function. Remark 13.
The construction in Proposition 12 may be generalized simply byvarying the exponent outside the finite product from 2 to any arbitrary integergreater than two. There then results an infinite family of such exceptional facto-rials. The quantity defined by Q n − j =1 ( n − j )! a , may be thought of as an abstractgeneralization of the super factorial (see [23], A000178) . Example 14.
The first few terms of the exceptional factorial defined in Propo-sition 12 are given by a = 1 , a = 3! a = 12 , a = 497664 , a = 6! a =443722221348087398400 , etc. Since the preceding results are valid for abstract factorials they include, in particu-lar, the recent factorial function considered in [3], and we summarize its constructionfor completeness. Let X ⊆ Z be a finite or infinite set of integers. Following [3], wedefine the notion of a p -ordering of X and use it to define the generalized factorialsof the set X inductively. By definition 0! X = 1. For p a prime, we fix an element a ∈ X and, for k ≥
1, we select a k such that the highest power of p dividing Q k − i =0 ( a k − a i ) is minimized. The resulting sequence of a i is called a p -orderingof X . As one can gather from the definition, such p -orderings are not unique, asone can vary a . Associated with such a p -ordering of X we define an associated p -sequence { ν k ( X, p ) } ∞ k =1 by ν k ( X, p ) = w p ( k − Y i =0 ( a k − a i )) , where w p ( a ) is, by definition, the highest power of p dividing a (e.g., w (162) = 81).It is then shown that although the p -ordering is not unique the associated p -sequence BSTRACT FACTORIALS 9 is independent of the p -ordering being used. Since this quantity is an invariant, onecan use this to define generalized factorials of X by setting k ! X = Y p ν k ( X, p ) , (5)where the (necessarily finite) product extends over all primes p . Example 15.
Bhargava’s factorial function (5) is an abstract factorial.
Hypothesis 1 of Definition 1 is clear by definition of the factorial in question. Hy-pothesis 2 of Definition 1 follows by the results in [3].As we mentioned above, the question of the possible existence of equal consecutivegeneralized factorials is of interest. We show herewith that such examples exist forabstract factorials over the ring of integers.
Example 16.
There exist sets X with consecutive equal Bhargava factorials, ! X .Perhaps the easiest example of such an occurrence lies in the set of generalizedfactorials of the set of cubes of the integers, X = { n : n ∈ N } , where one canshow directly that 3! X = 4! X (= 504). Actually, the first occurrence of this is forthe finite subset { , , , , , , , } .Another such set of consecutive equal generalized factorials is given by the finiteset of Fibonacci numbers X = { F , F , . . . , F } , where one can show directly that7! a = 8! a (= 443520). We point out that the calculation of factorials for finite setsas defined in [3] is greatly simplified through the use of Crabbe’s algorithm [6].Inspired by the factorial representation of the base of the natural logarithms, oneof the basic objects of study here is the series defined by the sum of the reciprocalsof the factorials in question. Definition 17.
For a given abstract factorial we define the constant e a by the seriesof reciprocals of its factorials, i.e., e a ≡ ∞ X r =0 r ! a . (6)Note that the series appearing in (6) converges on account of Definition 1(3) and1 < e a ≤ e . 3. Factorial sets and their properties
Besides creating abstract factorials using clever constructions, the easiest way togenerate them is by means of integer sequences. As we referred to earlier it isshown in [3] that on every subset X ⊆ Z + one can define an abstract factorial. Weshow below that there are other (non-unique) ways of generating abstract factorials(possibly infinitely many) out of a given set of positive integers. The construction of a factorial set.
Given I = { b , b , . . . } , I ⊂ Z ,( b i = 0), with or without repetitions, we associate to it another set X I = { B , B ,. . . , B n , . . . } of positive integers, termed simply a factorial set of I . In this case I is called a primitive (set) of X I .The elements of this factorial set X I are defined as follows: B = 1 by definition, B is (the absolute value of) an arbitrary but fixed element of I , say, B = | b | (so that the resulting factorial set X generally depends on the choice of b ). Next, B is the smallest (positive) number of the form b α b α (where the α i >
0) suchthat B | B . Hence B = | b b | . Next, B is defined as the smallest (positive)number of the form b α b α b α such that B B | B . Thus, B = | b b b | . Now, B is defined as that smallest (positive) number of the form Q k =1 b kα k such that B B | B and B | B . This calculation gives us B = | b b b b | . In general,we build up the elements B i , i = 2 , , . . . , n − , inductively as per the precedingconstruction and define the element B n as that smallest (positive) number of theform | Q nk =1 b kα k | such that B i B j | B n for every i, j , 0 ≤ i ≤ j ≤ n , and i + j = n . Remark 18.
Observe that permutations of the set I may lead to different factorialsets, X I , each one of which will be used to define a different abstract factorial(below).It is helpful to think of the elements B n of a factorial set as defining a sequence of generalized factorials . In [3] one finds that the set of ordinary factorials arises froma general construction applied to the set of positive integers. For the analogue ofthis result see Section 5.The basic properties of any one of the factorial sets of a set of integers, all of whichfollow from the construction, can be summarized as follows. Remark 19.
Let I = { b i } ⊂ Z be any infinite subset of non-zero integers. Forany fixed b m ∈ I , consider the (permuted) set I ′ = { b m , b , b , . . . , b m − , b m +1 , . . . } .Then the factorial set X b m = { B , B , . . . , B n , . . . } of I ′ exists and for every n > i, j ≥ i + j = n , we have B i B j | B n . In addition, if the elements of I are all positive, then the B i are monotone.Of course, factorial sets may be finite (e.g., if X is finite) or infinite. The nextresult shows that factorial sets may be used to construct infinitely many abstractfactorials. Theorem 20.
Let I be an integer sequence, X I = { B n } one of its factorial sets.Then, for each k ≥ , the map ! a : N → Z + defined by ( a = 1 ) n ! a = n ! B B · · · B n + k , is an abstract factorial on I . Varying k of course produces an infinite family of abstract factorials on I . Theabove construction of a factorial set leads to very specific sets of integers, setswhose elements we characterize next. (In the sequel, as usual, ⌊ x ⌋ is the greatestinteger not exceeding x .) BSTRACT FACTORIALS 11
Theorem 21.
Given I = { b i } ⊂ Z + , the terms B n = b ⌊ n ⌋ b ⌊ n/ ⌋ b ⌊ n/ ⌋ · · · b n ⌊ n/n ⌋ (7)characterizes one of its factorial sets, X b .The next result leads to a structure theorem for generalized binomial coefficientscorresponding to factorial functions induced by factorial sets. Proposition 22.
With B n defined as in (7) we have, for every n ∈ Z + and forevery k = 0 , , . . . , n , B n B k B n − k = n Y i =1 b iα i , α i = 0 or . (8)With this the next result is clear. Corollary 23.
Let n ! | B n for all n ∈ Z + . Then n ! a = B n is an abstract factorialwhose generalized binomial coefficients are of the form (cid:18) nk (cid:19) a = n Y i =1 b iα i , α i = 0 or . (9)4. Irrationality results
We now state a few lemmas leading to a general irrationality result for sums of re-ciprocals of abstract factorials. First we note that given a positive integer sequence b n the series ∞ X n =0 n ! b n (10)may be either rational or not. Indeed, Erd¨os [8] pointed out that the series ∞ X n =0 n !( n + 2) = 1 . The problem in this section consists in determining irrationality criteria for seriesof the form (10) using abstract factorials.
Lemma 24.
Let ! a be an abstract factorial whose factorials satisfy (4) . Then e a is irrational. Remark 25.
Although condition (3) in Definition 1 (i.e., n ! | n ! a ) of an abstractfactorial appears to be very stringent, one cannot do without something like it; thatis Lemma 24 above is false for generalized factorials not satisfying this or some othersimilar property. For example, for q > n ! a = q n .It satisfies properties (1) and (2) of Definition 1 but not (3). In this case it is easyto see that even though our function satisfies equation (4), e a so defined is rational. Corollary 26.
Let X be the set of prime numbers and ! a the factorial function [3] of this set given by [3] n ! a = Y p p P ∞ m =0 [ n − pm ( p − ] , (11) where the (finite) product extends over all primes. Then e a ≈ . is irra-tional. The previous result holds because the generalized factorials of the set of primessatisfy (4) with equality. The next lemma covers the logical alternative exhibitedby equation (3) in Lemma 10.
Lemma 27.
Let ! a be an abstract factorial whose factorials satisfy (3) . Then e a is irrational. Combining the previous two lemmas we find the following theorem,
Theorem 28.
For any abstract factorial, ! a the number e a is irrational. In fact,if ! a , ! a ,. . . , ! a k is any collection of factorial functions, s i ∈ N , i = 1 , , . . . , k , notall equal to zero, then ∞ X n =0 Q kj =1 n ! s j a j is irrational. The alternating series counterpart of the previous theorem is next.
Theorem 29.
Let ! a , ! a ,. . . , ! a k be any collection of factorial functions, s i ∈ N , i = 1 , , . . . , k , not all equal to zero. Then ∞ X n =0 ( − n Q kj =1 n ! s j a j is irrational. Remark 30.
This shows that the irrationality of the constants e a appears to bedue more to the structure of the abstract factorial in question than an underlyingtheory about the base of the natural logarithms. Example 31.
Let n ! a := (2 n )! / n , n = 0 , , . . . . Then this is an abstract factorial.An immediate application of Theorems 28 and 29 in the simplest case where s = 1gives that both quantities ∞ X n =0 n (2 n )! = cosh √ , and ∞ X n =0 ( − n n (2 n )! = cos √ BSTRACT FACTORIALS 13
More generally, for b ∈ Z + the quantity n ! b = ( bn )! /b n , 0! b = 1, defines an abstractfactorial so that generally, ∞ X n =0 b n ( bn )!is irrational as well along with its alternating series counterpart. Example 32.
Let F n denote the classical Fibonacci numbers defined by the recur-rence relation F n = F n − + F n − , F = F = 1. The “Fibonacci factorials” ([23],id:A003266), denoted here by F ( n ) are defined by F ( n ) = n Y k =1 F k . Define a factorial function by setting F (0) := 1 and n ! a := n ! F ( n ) , n = 1 , , . . . In this case, the generalized binomial coefficients involve the
Fibonomial coefficients (= F ( n ) / F ( k ) F ( n − k )) so that (cid:18) nk (cid:19) a = (cid:18) nk (cid:19) (cid:18) nk (cid:19) F where the Fibonomial coefficients on the right ([23], id:A010048), [13], are integersfor k = 0 , , . . . , n . Once again, an application of Theorem 28 yields that for every k ∈ Z + , ∞ X n =0 n !( F ( n )) k is irrational. Example 33.
The (exceptional) abstract factorial of Proposition 12 gives therapidly growing (irrational) series of reciprocals of factorials: e a = 1 + 1 + 112 + 112 + 1497664 + 1443722221348087398400 + · · · An inverse problem
We recall that a set X is called a primitive of an abstract factorial ! a if the sequenceof its factorials, { n ! a } ∞ n =0 coincides with one of the factorial sets of X . The questionwe ask here is: When does an abstract factorial admit a primitive set?
Firstly,we give a simple necessary and sufficient condition for the existence of such aprimitive set and, secondly, we give examples, the first of which shows that theordinary factorial function has a primitive set whose elements are simply given bythe exponential of the Von Mangoldt function.
Lemma 34.
A necessary and sufficient condition that a set X = { b n } be a primitiveof the abstract factorial n ! a is that the quantity b n = n ! a Q n − i =1 b ⌊ n/i ⌋ i (12) defined recursively starting with b = 1! a , be an integer for every n > . Remark 35.
It is not the case that (12) is always an integer even though the firstthree terms b , b , b are necessarily so. The reader may note that the abstractfactorial defined by n ! a = (2 n )! / n has no primitive since b = 14 /
3. On the otherhand, the determination of classes of abstract factorials that admit primitives is afascinating problem.We cite two examples of important factorials that do admit primitives.
Theorem 36.
The ordinary factorial function has for a primitive (besides the set Z + ) the set X = { b n } where b n = e Λ( n ) , where Λ( n ) is the Von Mangoldt function. Another example of an abstract factorial that admits a primitive (other than theoriginal set it was intended for) is the factorial function [3] for the set of primes. Inother words, there is a set X different from the set of primes whose factorials (asdefined herein) coincide with the abstract factorial n ! a = ( n + 1)! X = Y p p P ∞ m =0 [ npm ( p − ] , (13)obtained in [3] for the set of primes. (The factorial there is denoted by n ! X ).) Theorem 37.
The abstract factorial defined in (13) has for a primitive (besidesthe unordered set of prime numbers) the ordered set X = { b i } where here b = 2 ,and the remaining b n are given recursively by (12) and explicitly as follows: b i = , if i = p m ( p − for any prime p and any m ≥ , Y p, i = p m ( p − p, if i = p m ( p − for some prime p and m ≥ ,where the product extends over all primes p such that i has a representation in theform i = p m ( p − , for some m ≥ . The first few terms of the set X in Theorem 37 are given by X = { , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , . . . } It follows from Theorem 37 that if i is odd then b i = 1, necessarily. It is temptingto conjecture that every term in X is the product of at most two primes and thisis, in fact, true. Proposition 38.
Let n ∈ Z + . Then there are at most two (2) representations of n in the form n = p m ( p − where p is prime and m ∈ N . Applications
Before proceeding with some applications we require a few basic lemmas, the firstof which, not seemingly well-known, is actually due to Hermite ([12], p.316) andrediscovered a few times since. e.g., Basoco ([2], p.722, eq. (16).)
BSTRACT FACTORIALS 15
Lemma 39.
For k ≥ an integer, let σ k ( n ) denote the sum of the k -th powers ofthe divisors of n , (where, σ ( n ) = d ( n ) ). Then n X i =1 σ k ( i ) = n X i =1 i k ⌊ n/i ⌋ . (14) Note:
The left-side of (14) is the summatory function of σ k ( i ) or n times theaverage order of σ k ( i ) over its range ([11], Section 18.2). Furthermore, there is aninteresting relationship between (14) and the Riemann zeta function at the positiveintegers, that is, n X i =1 σ k ( i ) = ζ ( k + 1) k + 1 n k +1 + O ( n k )where the remainder terms are in terms of Ramanujan sums. Lemma 40.
Let α ( n ) denote the cumulative product of all the divisors of thenumbers , , . . . , n . Then α ( n ) = n Y i =1 i ⌊ n/i ⌋ . (15) Remark 41.
It is also known that α ( n ) = n Y k =1 ⌊ nk ⌋ !(see [23], id.A092143, Formula).We now move on to examples where we describe explicitly some of the factorial setsof various basic integer sequences. Example 42.
The factorial set X I of the set I of basically identical integers, I = { , q, q, q, q, . . . } as per our construction where q ≥
2, and B = q , gives thefactorial set X I = { , q, q , q , q , q , q , q , q , q , q , q , q , . . . } (16)a set whose n -th term is B n = q a ( n ) , where a ( n ) = P nk =1 d ( k ) (by Theorem 21and Lemma 39) and d ( k ) is, as before, the number of divisors of k . The functiondefined by setting n ! a = n ! B n defines an abstract factorial. Here we see that equalconsecutive factorials cannot occur by construction. In addition, by Lemma 24, ∞ X n =1 n ! q P nk =1 d ( k ) is irrational. Definition 43.
Let I be an infinite subset of Z + with a corresponding factorial set X I = { B n } . If n ! | B n for every n , we say that this factorial set X I is a self-factorialset .The motivation for this terminology is that the function defined by setting n ! a = B n is an abstract factorial. In other words, a self-factorial set may be thought of as aninfinite integer sequence of consecutive generalized factorials (identical to the set itself, up to permutations of its elements). The next result is very useful when onewishes to iterate the construction of a factorial set ad infinitum (i.e., when findingthe factorial set of a factorial set, etc.). Lemma 44. If I = { b n } is a set with n ! | b n for every n , then its factorial set X b is a self-factorial set. The same idea may be used to prove that
Corollary 45.
The factorial set X B of a self-factorial set X = { B n } is a self-factorial set. Next, we show that set Z + has a factorial set with interesting properties. Example 46.
We find a factorial set of the set X = Z + as per the precedingconstruction. Choosing B = 1 we get the following set, X Z + = { , , , , , , , , , , . . . } (17)a set which coincides (by Lemma 40 and Theorem 21) with the set of cumulativeproducts of all the divisors of the numbers 1 , , . . . , n (see Sloane [23], id.A092143).Note that by construction n ! | B n for every n . Hence, we can define an abstractfactorial by setting n ! a = B n to find that for this factorial function the set of facto-rials is given by the set itself , that is, this X Z + is self-factorial. In particular, equalconsecutive factorials cannot occur by construction, and it follows from Lemma 24that the number defined by the sum of the reciprocals of these B n , i.e., e a = 1 + ∞ X n =1 / n Y i =1 i ⌊ n/i ⌋ = 1 + 1 + 12 + 16 + 148 + 1240 + . . . ≈ . I have the prop-erty that n ! | B n for all n . Such sequences can thus be used to define abstractfactorials. For example, if we consider the set of all k -th powers of the integers, I = { n k : n ∈ Z + } , k ≥
2, then another application of Lemma 40 shows that itsfactorial set X I (with B = 1) is given by terms of the form B n = n Y i =1 i k ⌊ n/i ⌋ . In these cases we can always define an abstract factorial by writing n ! a = B n .Indeed, the semigroup property of abstract factorials (Proposition 3) implies thatfor each k ∈ Z + the series of k -th powers of the reciprocals of this cumulativeproduct, ∞ X n =1 / n Y i =1 i k ⌊ n/i ⌋ is irrational. Remark 47.
The previous results are a special case of a more general result whichstates that the factorial set of the set X = q Z + , q ∈ Z + , is given by terms of theform B n = q P nk =1 d ( k ) n Y i =1 i ⌊ n/i ⌋ . BSTRACT FACTORIALS 17
This is readily ascertained using the representation theorem, Theorem 21, andLemma 40.
Example 48.
Let q ∈ Z + , q ≥ X = { q n : n ∈ N } . Then thegeneralized factorials [3] of this set are given simply by n ! a = Q nk =1 ( q n − q k − ),[3]. The factorial set X q of this set X defined by setting B = q yields the set X q = { , q, q , q , q , q , q , q , q , q , q , q , . . . } , (18)whose n -th term is B n = q a ( n ) by Lemma 39, where a ( n ) = σ (1)+ . . . + σ ( n ) is (n-times) the average order of σ ( n ), ([11], Section 18.3, p.239, p. 266). The averageorder of the arithmetic function σ ( n ) is, in fact, the a ( n ) defined here, its asymp-totics appearing explicitly in ([11], Theorem 324). Note that this sequence a ( n )appears in ([23], id.A024916) and that n ! does not divide B n generally, so this setis not self-factorial. However, one may still define infinitely many other factorialson it as we have seen (see Theorem 20) . Example 49.
Let q ≥ X = { q n : n ∈ N } . The factorial set X q of this set X defined by setting B = q givesthe set X q = { , q, q , q , q , q , q , q , q , q , q , q , . . . } , (19)where now the n -th term is B n = q a ( n ) by Lemma 39, where a ( n ) = P nk =1 σ ( k )and σ ( k ) represents the sum of the squares of the divisors of k ([11], p.239).The previous result generalizes nicely. Example 50.
Let q ≥ k ≥ X = { q n k : n ∈ N } . In this case, the factorial set X q of this set X defined as usualby setting B = q gives the set whose n -th term is B n = q a k ( n ) by Lemma 40,where a k ( n ) = P ni =1 σ k ( i ) and σ k ( i ) is the sum of the k -th powers of the divisorsof i ([11], p.239).6.1. Factorial sets of the set of primes.
In this section we find a factorial setfor the set of primes that leads to a factorial function that is different from the onefound in [3] and describe a few of its properties.
Example 51.
Let I = { p i : i ∈ Z + } be the set of primes. Setting B = 2 weobtain the characterization of one of its factorial sets, i.e., X I = { , , , , , , , , . . . } in the form, X I = { B n } where (according to our construction), B n = 2 n ⌊ n/ ⌋ ⌊ n/ ⌋ · · · p i ⌊ n/i ⌋ · · · p n ⌊ n/n ⌋ = n Y i =1 p i ⌊ n/i ⌋ . (20)First we note that for each n the total number of prime factors of B n is equal to d (1) + d (2) + · · · + d ( n ). Next, this particular factorial set X is actually containedwithin a class of numbers considered earlier by Ramanujan [16], namely the classof numbers of the form Q ni =1 p ia i where a ≥ a ≥ . . . ≥ a n , a class which includesthe highly composite numbers (hcn) he had already defined in 1915. In addition, the superadditivity of the floor function and the representation of theordinary factorial function as a product over primes ([15], Theorem 27) shows thatfor every positive integer n , n ! | B n , where B n is as in (20) (we omit the details).This now allows us to define an abstract factorial by writing n ! a = B n . Since X I is a self-factorial set and there are no consecutive factorials we conclude fromLemma 24 that e a = 1 + ∞ X n =1 / { n ⌊ n/ ⌋ ⌊ n/ ⌋ · · · p i ⌊ n/i ⌋ · · · p n ⌊ n/n ⌋ } ≈ . , is irrational. The semigroup property of abstract factorials (Proposition 3) impliesthat the sum of the reciprocals of any fixed integer power of B n is irrational as well.The arithmetical nature of the generalized binomial coefficients (defined in Defini-tion 1(2)) corresponding to the abstract factorial (20) inspired by the set of primesis to be noted. It follows by Proposition 22 that Proposition 52.
The factorial function defined by n ! a = B n where B n is definedin (20) has the property that for every n and for every k, ≤ k ≤ n , the generalizedbinomial coefficient (cid:0) nk (cid:1) a is odd and square-free. Remark 53.
In 1980 Erd¨os and Graham [9] made the conjecture that the (or-dinary) central binomial coefficient (cid:0) nn (cid:1) is never square-free for n >
4. In 1985S´ark¨ozy [20] proved this for all sufficiently large n , a result that was extended laterby Sander [19]. Proposition 52 above implies the complementary result that the(generalized) central binomial coefficient (cid:0) nn (cid:1) a associated with the abstract factorialinduced by the set of primes (20) is always square free, for every n .Now, observe that the first 6 elements of our class X I (defined in Example 51) arehcn; there is, however, little hope of finding many more due to the following result. Proposition 54.
The sequence defined by (20) contains only finitely many hcn.
Remark 55.
It is interesting to note that the first failure of the left side of (37)in the proof of this result is when n = 9. Comparing all smaller hcn (i.e., thosewith a ≤
8) with our sequence we see that there are no others (for a table of hcnsee [18] (pp.151-152)); thus the 6 found at the beginning of the sequence are theonly ones. The sequence B n found here grows fairly rapidly: B n ≥ n +1 p p · · · p n although this is by no means precise.Actually more is true regarding Proposition 54. The next result shows that hcn arereally elusive . . . Proposition 56.
The integer sequences defined by taking any of our factorial set(s)of the set of primes, even factorial sets of the factorial sets of the set of primes etc.contain only finitely many hcn.
On factorials of the primes and abstract factorials.
We consider herethe question of whether the set of the factorials of the primes is an abstract factorial.
BSTRACT FACTORIALS 19
To be precise, define f : N → Z + as follows: f ( n ) = , if n = 0 , , if n = 1 .p n − ! , if n ≥ . The question we ask is whether f is an abstract factorial? The answer seems farfrom obvious. A numerical search seems to indicate that the first few binomialcoefficients are indeed integers (at least up to n = 50). Indeed, use of the lowerbound [7] p n − > ( n − { log( n −
1) + log log( n − − } for all n ≥ n − { log( n −
1) + log log( n − − } − n > n (by elementary Calculus) so that p n − > n for all n ≥
7. We concludethat n ! | f ( n ), for all n .Now consider the (abstract) binomial coefficients (cid:18) n + 1 k + 1 (cid:19) a = p n ! p k ! p n − k − !where we can assume, without loss of generality, that n ≥ ≤ k ≤ n − n > k we factor out p k !from the numerator thereby leaving a product of p n − p k consecutive integers thatare necessarily divisible by ( p n − p k )!. Thus, in order to prove that these abstractbinomial coefficients are indeed integers it suffices to show that p n ≥ p k + p n − k − , ≤ k ≤ n − , (21)and all n ≥ π ( x ), i.e., π ( x + y ) ≤ π ( x ) + π ( y )for all x, y ≥ p n ≥ p n − k + p k +1 − ≤ k ≤ ( n − / n ≥
3, a conjecture that has not been settled yet. However,since p k +1 > p k + 1 and p n − k > p n − k − it follows that (22) implies (21). So,any counterexample to (21) also serves as a counterexample to the stated Hardy-Littlewood conjecture. Still, (21) may be true, i.e., f ( n ) is an abstract factorial.However, settling (21) one way or another is beyond the scope of this work.6.3. Factorial sets of sets of highly composite numbers.
It turns out thatthere are hcn that are divisible by arbitrarily large (ordinary) factorials.
Proposition 57.
Let m ∈ Z + . Then there exists a highly composite number N such that m ! | N . Remark 58.
It is difficult to expect Proposition 57 to be true for all hcn largerthan N as can be seen by considering the hcn N = 48 where 4! |
48 but 4! does notdivide the next hcn, namely, 60. However, the proof shows that Proposition 57 istrue for all those hcn larger than N for which the largest prime p (appearing in theprime factorization of N ) both exceeds e m and appears in subsequents hcn’s primefactorization. (This is, of course, not always the case: e.g., the largest prime in theprime decomposition of 27720 is 11 but the largest such prime for the next hcn,namely 45360, is 7.) Terminology:
We will denote by H = { h n } a collection of hcn with the propertythat n ! | h n for each n ∈ Z + (note that the existence of such a set is guaranteed byProposition 57). Proposition 59.
The factorial set H h of H is self-factorial and for k ≥ theseries of the reciprocals of various powers of these hcn, i.e., ∞ X n =1 / { h ⌊ n/ ⌋ h ⌊ n/ ⌋ h ⌊ n/ ⌋ · · · h ⌊ n/n ⌋ n } k is irrational. Final Remarks
We add a few remarks about further irrationality results and integral representa-tions of abstract factorials using as a basis, the Stieltjes moment problem. Forbackground material we refer the reader to either Akhiezer [1] or Simon [22].We state the Stieltjes moment problem for completeness: Given a sequence s , s , . . . of real numbers to determine a measure dψ on [0 , ∞ ) such that for every n ≥ s n = Z ∞ x n dψ ( x ) . If there is one, to determine if and when it is unique and how it can be generated.One of the basic results in this area is a theorem of Carleman [5] which states thatthe Stieltjes moment problem has a unique solution (i.e., is determinate ) providedthe moments satisfy the criterion, ∞ X n =1 s n − / n = + ∞ . In our case we consider those (necessarily positive) sequences s n generated by ab-stract factorials, n ! a , for n ≥
0. The prototype here is the ordinary factorial, n !, where dψ ( x ) = e − x dx , which gives the classic relation between factorials andEuler’s Gamma function, n ! = Z ∞ x n e − x dx. Another less obvious example arises from a study of the abstract factorial in Ex-ample 46. For any given a ∈ Z + , the abstract factorial n ! a = ( an )! /a n may be BSTRACT FACTORIALS 21 represented as ( an )! a n = Z ∞ x n dψ ( x ) , where dψ ( x ) = exp {− ( ax ) /a } ( ax ) − /a dx. Here we note that each of these measures dψ ( x ) are also unique by the stated resultof Carleman.The case of a general abstract factorial n ! a is much more difficult. Even though weknow there exists a function φ of bounded variation and of finite total variation on[0 , ∞ ) such that n ! a = Z ∞ x n dφ ( x ) , see, e.g., [4], the problem is whether this φ is unique let alone exhibiting such afunction in this generality.Integral representations of abstract factorials in terms of a solution of a Stieltjesmoment problem may be useful in the search for transcendence proofs for the variousirrational numbers encountered here using the ideas buried in Hilbert’s (1893) proofof the transcendence of e using the Gamma function.To get irrationality results of the type presented here it merely suffices to have atour disposal an abstract factorial, as then this factorial function will provide thedefinition of a self-factorial set. For example, the following sample theorems are aneasy consequence of Theorem 28 and the other results herein. Theorem 60.
Let q n ∈ Z + be a given integer sequence satisfying q = 1 and forevery n ≥ , q i q j | q n for all i, j , ≤ i, j ≤ n with i + j = n . Then the series ∞ X n =0 n ! q n is irrational. Corollary 61.
Let f : Z + × Z + → Z + and let f ( · , q ) be concave for each q ∈ Z + .Then, for any q ∈ Z + , ∞ X n =1 q f ( n,q ) n ! is irrational. In fact, binomial coefficients can, in some cases, be used to induce abstract factorialsas well as one can gather from the following consequence of the previous theorem.
Corollary 62.
Let q ∈ Z + . Then ∞ X n =1 /n ! q ( n + q − q ) is irrational. (The alternate series counterpart of the preceding result is also irrational as usual.) Theorem 63.
Let q ∈ Z + . Then both ∞ X n =1 n ! qn ∞ X n =1 ( − n n ! qn are irrational. Example 64.
Let q n = n ! / ⌊ n/ ⌋ , n = 0 , , , . . . . Then q n ∈ Z + for every n , n ! q n is an abstract factorial and a straightforward calculation gives us that q i q j | q n forall i, j , ≤ i ≤ j ≤ n with i + j = n . Hence the series ∞ X n =0 ⌊ n/ ⌋ n ! = 14 (1 + √ I o ( √ − √ J o ( √ ≈ . . . . is irrational. (Here I o , J o are Bessel functions of the first kind of order .) As a final result we show independently that
Theorem 65. ∞ X n =1 p n ! is irrational. If our function f , defined earlier, (basically the n -th prime factorial function) turnsout being an abstract factorial this would not only lead to Theorem 65 immediatelybut also generate other such irrationality results using products of the factorials ofthe n -th prime with other abstract factorials, as we have seen.8. Proofs
Proof. (Proposition 3) It suffices to prove this for any pair of abstract factorials.To this end, write n ! a = n ! · n ! where n ! i , i = 1 , a = 1, and n ! | n ! a . Finally, observe that (cid:18) nk (cid:19) a = Y i =1 (cid:18) nk (cid:19) i where, by hypothesis,each binomial coefficient on the right has integral values. (cid:3) Proof. (Proposition 4) Note that (1) implies that the generalized binomial coeffi-cients of the factorial n ! a = n ! h n are integers. In addition, since h = 1, 0! a = 1,the divisibility condition is clear. The converse is also clear and so is omitted. (cid:3) Proof. (Lemma 8) Assuming the contrary we let ! a be such a factorial and let k ≥ r k = r k +1 = 1. Since the binomial coefficient (cid:18) k + 2 k (cid:19) a = ( k + 2)! a a k ! a = 12! a ∈ Z + , by Definition 1(2), this implies that 2! a | k . On the other hand, 2! | a byDefinition 1(3), so we get a contradiction. (cid:3) BSTRACT FACTORIALS 23
Proof. (Lemma 9) Lemma 8 guarantees that r k − = 1. Hence r k − ≥
2. Assume,if possible, that r k − = 2. Since ( k + 1)! a = k ! a = 2( k − a and the generalizedbinomial coefficient (cid:18) k + 1 k − (cid:19) a = ( k + 1)! a a ( k − a = 22! a is a positive integer, 2! a must be equal to either 1 or 2. Hence, by hypothesis, itmust be equal to 1. But then by Definition 1(3) 2! must divide 2! a = 1, so we geta contradiction. (cid:3) Proof. (Lemma 10) The sequence of factorials n ! a is non-decreasing by Remark 2,thus, in any case lim sup k →∞ /r k ≤ . Next, let k n ∈ Z + , be a given infinitesequence. There are then two possibilities: Either there is a subsequence, denotedagain by k n , such that k n ! a = ( k n + 1)! a for infinitely many n , or every subsequence k n has the property that k n ! a = ( k n + 1)! a except for finitely many n . In the firstcase we get (3). In the second case, since k n ! a divides ( k n + 1)! a (by Definition 1)it follows that ( k n + 1)! a ≥ k n ! a , except for finitely many n and this now implies (4).The final statement is supported by an example wherein X is the set of all (ordinary)primes, and the factorial function is in the sense of [3]. In this case, the explicitformula derived in ([3], p.793) for these factorials can be used to show sharpnesswhen the indices in (4) are odd, since then r k = 2 for all such k . (See the proof ofCorollary 26 below.) (cid:3) Proof. (Proposition 12) To see that this is a factorial function we must show thatthe generalized binomial coefficients (cid:0) nk (cid:1) a are positive integers for 0 ≤ k ≤ n as theother two conditions in Definition 1 are clear by construction. Putting aside thetrivial cases where k = 0 , k = n we may assume that 1 ≤ k ≤ n − (cid:0) nk (cid:1) a ∈ Z + for k = 1 , , . . . , n − n ! a necessarily contains two copies of each of the terms k ! a and( n − k )! a for each such k whenever 2 k = n . It follows that the stated binomialcoefficients are integers whenever 2 k = n . On the other hand, if 2 k = n thetwo copies of k ! a in the denominator are canceled by two of the respective fourcopies in the numerator (since now ( n − k )! a = k ! a ). Observe that (3) holds byconstruction. (cid:3) Proof. (Theorem 20) One need only apply the Definition of an abstract factorialand the construction of the B n of this section. The only part that needs a minorexplanation is the integer nature of the generalized binomial coefficients. However,note that for fixed k ∈ N , (cid:18) nr (cid:19) a = (cid:18) nr (cid:19) n + k − r Y i =1 B r + i B i , where 1 ≤ r ≤ n −
1, the other cases being trivial. Finally, the right hand side mustbe an integer since each ratio B r + i /B i is also an integer, by construction. (cid:3) Proof. (Theorem 21) Note that (7) holds for the first few n by inspection so we usean induction argument: Assume that B i = i Y j =1 b j ⌊ i/k ⌋ holds for all i ≤ n −
1. Since we require B i B j | B n for every i, j , 0 ≤ i ≤ j ≤ n and i + j = n , we note that B i B n − i | B n for i = 0 , , . . . , ⌊ n/ ⌋ . On the otherhand if this last relation holds for all such i then by the symmetry of the productinvolved we get B i B j | B n for every i, j , 0 ≤ i ≤ j ≤ n and i + j = n . Now, writing B n = b α b α · · · b nα n where the α i > B i B n − i , that is B i B n − i = i Y j =1 b j ⌊ i/j ⌋ n − i Y j =1 b j ⌊ ( n − i ) /j ⌋ , = i Y j =1 b j ⌊ i/j ⌋ + ⌊ ( n − i ) /j ⌋ n − i Y j = i +1 b j ⌊ ( n − i ) /j ⌋ . where i ≤ ( n − /
2. Comparison of the first and last terms of this product withthe expression for B n reveals that α = n and α n = 1. For a given j , 1 ≤ j ≤ n ourconstruction and the induction hypothesis implies that α i = 1 + ⌊ ( n − i ) /i ⌋ = ⌊ n/i ⌋ since B i B n − i | B n . This completes the induction argument. (cid:3) Proof. (Proposition 22) Set aside the cases k = 0 , n as trivial. Since B n /B k B n − k = B n /B n − k B k we may assume without loss of generality that k ≥ n/ n ≥
2. Using the expression (7) for B n we note that the left hand side of (8) maybe rewritten in the form, B n B k B n − k = n − k Y j =1 b j ⌊ n/j ⌋−⌊ k/j ⌋−⌊ ( n − k ) /j ⌋ · k +1 Y j = n − k +1 b j ⌊ n/j ⌋−⌊ k/j ⌋ · n Y j = k +2 b j ⌊ n/j ⌋ . (23)Now the first term in the first product must be 1 since j = 1 and n, k are integers.Next, since ⌊ x + y ⌋ ≤ ⌊ x ⌋ + ⌊ y ⌋ + 1 , for all x, y ≥
0, replacing x by x − y we get0 ≤ ⌊ x ⌋ − ⌊ y ⌋ ≤ ⌊ x − y ⌋ , x ≥ y. Hence those exponents corresponding to j ≥ ⌊ ( n − k ) /j ⌋ = 0 over the range j = n − k + 1 , . . . , k + 1. Using this in the above display gives that the exponents inthe second product are bounded above by 1. The exponents in the third product arebounded above by ⌊ n/ ( k + 2) ⌋ ≤
1, they are non-increasing, and bounded below by1. Hence the exponents in the third product are all equal to 1. Thus the exponentsin (8) are either 0 or 1.The precise determination of the exponents in (23) is not difficult. For a given j ,whether 1 ≤ j ≤ n − k or n − k + 1 ≤ j ≤ k + 1, writing n, k in base j in the form n = n + n j + n j + . . . , etc. we see that, α j = , if n − k ≥ , , if n − k < , BSTRACT FACTORIALS 25
These results can be interpreted in terms of “carries” across the radix point ifrequired (see e.g., [13]). Finally the value α j = 1 in the range k + 2 ≤ j ≤ n . (cid:3) Proof. (Corollary 23) The assumptions imply that the generalized binomial coeffi-cients of the factorial defined here are given by the left side of (8). (cid:3)
Proof. (Lemma 24) The quantity 0! a = 1 by definition, so we leave it out of thefollowing discussion. Assume, on the contrary, that e a is rational, that is, E a ≡ e a − E a = a/b , for some a, b ∈ Z + , ( a, b ) = 1. In addition, E a − k X m =1 m ! a = ∞ X m = k +1 m ! a . Let k ≥ b , k ∈ Z + and define a number α k by setting α k ≡ k ! a E a − k X m =1 m ! a ! = k ! a ab − k X m =1 m ! a ! . (24)Since k ≥ b and k ! divides k ! a (by Definition 1(3)) it follows that b divides k ! a (since b divides k ! by our choice of k ). Hence k ! a a/b ∈ Z + . Next, for 1 ≤ m ≤ k we have that k ! a /m ! a ∈ Z + (by Definition 1(2)). Thus, α k ∈ Z + , for (any) k ≥ b .Note that, α k = k ! a ∞ X m = k +1 m ! a = k ! a (cid:18) k + 1)! a + 1( k + 2)! a + . . . (cid:19) . (25)First, we assume that L < /
2. For ε > L + ε < /
2, we choose N sufficiently large so that for every k ≥ N we have k ! a / ( k + 1)! a < L + ε. Then it iseasily verified that k ! a ( k + i )! a < ( L + ε ) i , for every i ≥ k ≥ N . Since L + ε < / α k ≤ ( L + ε ) ∞ X i =0 ( L + ε ) i = L + ε − ( L + ε ) < , and this leads to a contradiction.The case L = 1 / ε >
0, there exists an N such that for all k ≥ N , k ! a ( k + 1)! a ≤ / ε. Hence, for all k ≥ N , α k ≤ (1 / ε ) ∞ X i =0 (1 / ε ) i = 1 / ε − (1 / ε ) . (26) We now fix some ε < / N . Then the right-side of (26) isless than two. But for k ≥ N ≡ max { b, N } , α k is a positive integer. It followsthat α k = 1. Using this in (25) we get that for every k ≥ N ,1 = k ! a (cid:18) k + 1)! a + 1( k + 2)! a + . . . (cid:19) . (27)Since the same argument gives that α k +1 = 1, i.e.,1 = ( k + 1)! a (cid:18) k + 2)! a + 1( k + 3)! a + . . . (cid:19) , (28)comparing (27) and (28) we arrive at the relation ( k + 1)! a = 2 k ! a , for every k ≥ N . Iterating this we find that, under the assumption of equality in (4) wehave ( k + i )! a = 2 i k ! a , for each i ≥
1, and for all sufficiently large k . However,by Definition 1(3), ( k + i )! a = n k ! a i ! a for some n i ∈ Z + . Hence, n i i ! a = 2 i , forevery i , for some integer n i depending on i . This, however, is impossible since, byDefinition 1(4), i ! must divide i ! a . Thus, i ! must also divide 2 i for every i which isimpossible. This contradiction proves the theorem. (cid:3) Proof. (Corollary 26) The prime factorization of this factorial function is given inits definition, (11). Replacing n , now assumed odd, by n + 1, we see that the onlycontribution to ( n + 1)! X comes from an additional factor of 2, so that whenever n is odd, we have for these factorials for the set of primes X in [3], n ! X ( n + 1)! X = 12 . It now follows that (4) is satisfied, with equality, and so by Lemma 24, e a is irra-tional. (cid:3) Proof. (Lemma 27) Since 2! | a , 2! a must be even. There are now two cases: either2! a = 2 and this implies 2! a ≥ a = 2. Case 1:
Let 2! a = 2. We proceed as in the preceding Lemma 24 up to (25). Thusthe assumption that e a − e a − a/b implies that α k ∈ Z + (25) forany k ≥ b . So, α k = k ! a ∞ X n = k +1 n ! a = k ! a (cid:18) k + 1)! a + 1( k + 2)! a + . . . (cid:19) , = 1 /r k + 1 /r k r k +1 + ∞ X n =3 /r k r k +1 r k +2 · · · r k + n − , (29)Since the factorials have integral valued binomial coefficients we see that the product r r · · · r n − = n ! a / a is a positive integer for every n . Hence, (cid:0) n + kk (cid:1) a ∈ Z + isequivalent to saying that n ! a | r k r k +1 · · · r k + n − , for every k ≥ n ≥
1. Since n ! | n ! a for all n by Definition 1(3), this means that n ! | r k r k +1 · · · r k + n − , (30)for every integer k ≥ n ≥ BSTRACT FACTORIALS 27
By hypothesis there is an infinite sequence of equal consecutive factorials. There-fore, we can choose k sufficiently large so that k ≥ b and r k +1 = 1. Then (29) issatisfied for our k with the α k there being a positive integer. With such a k atour disposal, we now use Lemma 9 which forces r k ≥ a = 2). Using thisinformation along with (30) in (29) we get α k ≤ / / ∞ X n =3 /r k r k +1 r k +2 · · · r k + n − , ≤ / ∞ X n =3 /n ! ≤ / e − − / ≈ . ... and this yields a contradiction. Case 2:
Let 2! a = 2. We proceed as in Case 1 up to (29) and then (30) withoutany changes. Once again, we choose k ≥ b and r k +1 = 1. Since 2 = 2! a | r k r k +1 , wesee that r k must be a multiple of two. If r k = 2, then (29)-(30) together give theestimate α k ≤ / / e − − / ≈ . ... . However, since α k is a positiveinteger, we must have α k = 1 . Hence r k = 2 is impossible since the right side of(29) must be greater than 1. Thus, r k ≥
4. Now using this estimate once again in(29) we see that1 = α k ≤ / / ∞ X n =3 /r k r k +1 r k +2 · · · r k + n − , (31) ≤ / e − − / ≈ . ... (32)and there arises another final contradiction. Hence e a is irrational. (cid:3) Proof. (Theorem 28) This is an immediate consequence of Lemma 24, Lemma 27,and the semigroup property. (cid:3)
Proof. (Theorem 29) It suffices to prove this in the case of one factorial functionwith s = 1 (by the semigroup property). This proof is simpler than the previousproof of Lemma 24 in the unsigned case as it can be modeled on Fourier’s proof ofthe equivalent result for the usual factorial. On the assumption that the series hasa rational limit a/b , we let k > b and then define the quantity α k by α k = (cid:12)(cid:12)(cid:12)(cid:12) k ! a ab − k ! a k X m =0 ( − m m ! a (cid:12)(cid:12)(cid:12)(cid:12) . Arguing as in Lemma 24 we get that α k ∈ Z + for all sufficiently large k .Since the series is alternating and the sequence of factorials is non-decreasing itfollows by the theory of alternating series that0 < (cid:12)(cid:12)(cid:12)(cid:12) ab − k X m =0 ( − m m ! a (cid:12)(cid:12)(cid:12)(cid:12) < k + 1)! a . Combining the last two displays we obtain that for all sufficiently large k ,0 < α k < k ! a ( k + 1)! a , and this leads to an immediate contradiction if the factorials satisfy the alternative(4) in Lemma 10.On the other hand, if the factorials satisfy the alternative (3) then r k +1 = 1 forinfinitely many k . We proceed as in the proof of Lemma 27 above with minorchanges. Thus, assuming the series has a rational limit a/b , with a, b >
0, we canchoose k so large that k > b so that β k ≡ k ! a ∞ X m = k +1 ( − m m ! a = k ! a (cid:18) ( − k +1 ( k + 1)! a + ( − k +2 ( k + 2)! a + . . . (cid:19) , = ( − k +1 /r k + ( − k +2 /r k r k +1 ++ ∞ X n =3 ( − k + n /r k r k +1 r k +2 · · · r k + n − . (33)But β k is an integer by our choice of k . If now 2! a = 2 (Case 1), r k +1 = 1 impliesthat r k ≥ | β k | ≤ / / ∞ X n =3 /r k r k +1 r k +2 · · · r k + n − ≤ · · · ≤ . ... which gives a contradiction.On the other hand, if 2! a = 2 (Case 2) then r k +1 = 1 gives us that the first twoterms in (33) cancel out (regardless of the value of r k ). Hence | β k | ≤ ∞ X n =3 /r k r k +1 r k +2 · · · r k + n − ≤ ∞ X n =3 /n ! < , another contradiction. This completes the proof. (cid:3) Proof. (Lemma 34) According to Theorem 21 any primitive set of the given factorialhas elements B n of the form (7) for an appropriate choice of b i . Thus, if the givenfactorial has a primitive, then n ! a = B n for all n . This is the case if and only if the b n are given recursively by (12).Conversely, if { b i } ∞ i =1 is a set of integers satisfying the divisibility condition (12),then the set X = { b , b , . . . } is a primitive of this factorial. (cid:3) Proof. (Theorem 36) We define b = 1, b i = e Λ( i ) . The standard representation ofthe ordinary factorial as a product of primes ([11], Theorem 416) gives us thatlog n ! = X m ≥ ⌊ np m ⌋ log p = n X i =1 ⌊ ni ⌋ log Λ( i ) = log n Y i =1 b i ⌊ ni ⌋ = log B n . An application of Lemma 34 and (7) now gives the conclusion. (cid:3)
BSTRACT FACTORIALS 29
Proof. (Theorem 37) The b i being explicit, the proof is simply a matter of verifi-cation. Since (13) is to be equal to (7) it suffices to express the b i as products ofvarious primes subject to their definition in the statement of the theorem. Clearly, b = 2.We observe that whenever i = p m ( p −
1) for any prime p and any m ≥ ⌊ n/i ⌋ cannot appear as a summand in (13). Consequently, weset b i = 1 in that case (as we don’t want any contribution from such a term to(13)).This leaves us with integers i , 1 ≤ i ≤ n , that can be represented in the form i = p m ( p −
1) for some prime(s) and some m ≥ m ’s depending on p ofcourse).We fix i . It is not hard to verify that for a given m there can be at most one prime p such that i = p m ( p − i , there is a correspondingset of primes, say, π , π , . . . , π s , and corresponding exponents m , m , . . . , m s ≥ i = π m j j ( π j − j = 1 , , . . . , s . (For example, 4 = 2 (2 −
1) = 5 (5 − π = 2 , π = 5, m = 2 and m = 0.)We claim that b i = π π · · · π s . (Recall that i is fixed here.) Consider the term b ⌊ ni ⌋ i appearing in (7). Since b ⌊ ni ⌋ i = s Y j =1 π ⌊ ni ⌋ j = s Y j =1 π (cid:22) nπ m j j ( π j − (cid:23) j , and each multiplicand in the product must appear exactly once in the factorization(13), we must have all the terms in (13) accounted for.For if there is a prime say, π e , from (13) that is “left out” of the resulting expression(7), there must be a corresponding denominator in the sum appearing in (13) andso an integer i having a representation in the form π m e e ( π e − π e must then appear in the resulting definition of the corresponding b i . Thus allprimes appearing in (13) are accounted for in the expression (7) and so the twoquantities (13) and (7) must be equal. (cid:3) Proof. (Proposition 38) First note that there are (infinitely many) integers n that cannot be represented in the form n = p m ( p −
1) (34)where p is a prime and m ≥ n satisfy (34) for some pair p, m as required. Then, in any case, it isnecessary that p ≤ n + 1. So, either m = 0, or m > m = 0. Then n = p −
1, and conversely if n is of the form n = p − m = 0). Now assume that n admits anotherrepresentation in the form n = q m ( q −
1) where q = p is another prime and m > We claim that q is the largest prime factor of n . For otherwise, if P > q is thelargest such prime factor, then for some r >
0, there holds n = P r α = q m ( q − α ∈ Z + and ( P, α ) = 1. But since (
P, q ) = 1, it is necessary that P r | ( q − P > q this is impossible, of course. We have thus shown that if (34) holdswith a prime q and m > q is the largest prime factor of n . This then givesus another possible representation of n in the desired form, making this a total ofat most two representations.Let n be not of the form one less than a prime , or equivalently, m > n in this form. Fix such a representation (34). Arguing as inthe preceding case we deduce that p must be the largest prime factor of n . Becauseof this, we conclude that there can be no other representation.Thus, in conclusion, there are either no representations of a positive integer n inthe form (34) where p is a prime and m ≥ n = 7 , , etc.), there is only onesuch representation (e.g., n = 20 , , etc.) or there are two such representations(e.g., n = 4 , , etc.). (cid:3) Proof. (Lemma 39) In this generality this result is hard to find in the literature.The case k = 0 can be found in ([11], Theorem 320), while the case k = 1 is referredto in ([23], A024916). The general case can actually be found in either Basoco ([2],eq.(16)) or Hermite ([12], p. 316). (cid:3) Proof. (Lemma 40) Write down the list of all the divisors from 1 to n inclusively.For a given i , 1 ≤ i ≤ n , there are ⌊ n/i ⌋ multiples of the number i less than or equalto n . Hence i ⌊ n/i ⌋ divides our cumulative product by definition of the latter. Takingthe product over all integers i shows that Q ni =1 i ⌊ n/i ⌋ | α ( n ). But all the divisors of α ( n ) must also be in the list and so each must be a divisor of Q ni =1 i ⌊ n/i ⌋ , sincethere can be no omissions by the sieving method. The result follows. (cid:3) Proof. (Lemma 44) For let X b = { B n } be one of its factorial sets. By Theorem 21its terms are necessarily of the form B n = b ⌊ n ⌋ b ⌊ n/ ⌋ b ⌊ n/ ⌋ · · · b n ⌊ n/n ⌋ . Since n ! | b n by hypothesis it follows that n ! | B n as well, for all n , and so this setis self-factorial. If b is replaced by any other element of I , then it is easy to seethat n ! | B n once again as all the exponents in the decomposition of B n are at leastone. (cid:3) Proof. (Corollary 45) Since X is self-factorial, n ! | B n for all n . The elements B ′ n of X B are necessarily of the form B ′ n = B ⌊ n ⌋ B ⌊ n/ ⌋ B ⌊ n/ ⌋ · · · B n ⌊ n/n ⌋ . Hence, n ! | B ′ n for all n , and this completes the proof. (cid:3) Proof. (Proposition 52) The square free part is clear on account of Proposition 22and the fact that the b i are primes in the representation (7). That the binomial BSTRACT FACTORIALS 31 coefficients must be odd is also clear since all powers of 2 cancel out exactly by(20). (cid:3)
Proof. (Proposition 54) This uses a deep result by Ramanujan [16] on the structureof hcn. Once it is known that every hcn is of the form q ≡ a a a · · · p a p (35)where a ≥ a ≥ a ≥ · · · ≥ a p ≥ ⌊ log p log λ ⌋ ≤ a λ ≤ ⌊ log P log λ ⌋ , (36)for every prime index λ , ([16], III.6-10, eq.(54)), where P is the first prime after p .Now set λ = 2 in (36) and use the fact that for the n -th term, B n , the multiplicityof the prime 2 is n , i.e., a = n . Since p = p n by the structure theorem for B n , wehave P = p n +1 . Since p n = O( n log n ) for n >
1, ([15], Theorem 113), the rightside of (36) now shows that n ≤ ⌊ log p n +1 log 2 ⌋ = O(log( n )) + O(log log( n )) , (37)which is impossible for infinitely many n . The result follows. (cid:3) Proof. (Proposition 56) Let X = { p n } be the set of primes. Recall that a factorialset is defined uniquely once we fix a value for b , some element of X . The choice b = 2,. . . , b n = p n leads to the factorial set already discussed in Proposition 54.On the other hand, if b = 2 then B n can never be highly composite for any n bythe structure theorem for hcn. We now consider the factorial set X of X (itselfthe (main) factorial set of X defined by setting b = p = 2 and whose elementsare given by (20)). The elements of X are necessarily of the form B n, = B n B ⌊ n/ ⌋ B ⌊ n/ ⌋ · · · B n ⌊ n/n ⌋ , = p n ( p p ) ⌊ n/ ⌋ ( p p p ) ⌊ n/ ⌋ · · · ( p n p ⌊ n/ ⌋ p ⌊ n/ ⌋ · · · p ⌊ n/n ⌋ n ) ⌊ n/n ⌋ , = p P ni =1 ⌊ i/ ⌋ ⌊ n/i ⌋ p P ni =1 ⌊ i/ ⌋ ⌊ n/i ⌋ · · · p n P ni =1 ⌊ i/n ⌋ ⌊ n/i ⌋ , = p P ni =1 σ ( i ) · · · p n , where σ ( i ) is the sum of the divisors of i (see Lemma 39). The assumption that forsome n , B n, is a hcn leads to the estimate (see (36)) ⌊ log p n / log 2 ⌋ ≤ n X i =1 σ ( i ) ≤ ⌊ log p n +1 / log 2 ⌋ . (38)However, by Theorem 324 in [11], P ni =1 σ ( i ) = n π /
12 + O( n log n ) . On the otherhand, the right side of (38) is O(log n ) + O(log log n ). It follows that the right handinequality in (38) cannot hold for infinitely many n , hence there can only be finitelymany hcn in X .Observe that the more iterations we make on the factorial sets X , X , . . . , X k ,the higher the order of the multiplicity of the prime 2 in the decomposition of therespective terms B n,k , and this estimate cannot be compensated by the right sideof an equation of the form (38). (cid:3) Proof. (Proposition 57) Since each prime must appear in the prime factorization ofan hcn (when written as an increasing sequence) there exists a hcn of the form N = 2 a a a · · · p a p with p > e m ( e = 2 . ... ). Using the representation of the factorials as a productover primes we observe that m ! | N ⇐⇒ a λ ≥ X j ≥ ⌊ m/λ j ⌋ , for every λ , where λ = 2 , , , . . . , p . In order to prove the latter we note that (36)implies that it is sufficient to demonstrate that ⌊ log p log λ ⌋ ≥ X j ≥ ⌊ m/λ j ⌋ , or since p > e m by hypothesis, that it is sufficient to show that ⌊ m log λ ⌋ ≥ X j ≥ ⌊ m/λ j ⌋ , for every prime λ = 2 , . . . , p . The latter, however is true on account of theestimates ⌊ m log λ ⌋ ≥ mλ − X j ≥ m/λ j ≥ X j ≥ ⌊ m/λ j ⌋ , valid for all primes λ = 2 , , . . . , p . This completes the proof. (cid:3) Proof. (Proposition 59) Fix a factorial set H = { h n } . Then H contains termsof the form B n = Q nj =1 h j ⌊ n/j ⌋ by construction where the h i are hcn in H . Since n ! | h n Lemma 44 implies that the factorial set H is self-factorial. The conclusionabout the irrationality now follows by Theorem 28 since n ! a = B n defines a factorialfunction by construction of the respective factorial sets. (cid:3) Proof. (Theorem 60) The assumptions imply that n ! a = n ! q n is an abstract factorialso Theorem 28 applies and the result follows. (cid:3) Proof. (Corollary 61) Fix q ∈ Z + , q ≥
2. We define q = 1 and q n = q f ( n,q ) for n ≥
1. We need only verify the that the generalized binomial coefficients areintegers. This, however, is a consequence of the fact that, for any i , 1 ≤ i ≤ n , and i + j = n , q n q i q j = q f ( n,q ) − f ( i,q ) − f ( n − i,q ) , along with the concavity of f in its first variable. The result is now a consequenceof Theorem 60. (cid:3) Proof. (Corollary 62) Fix q ∈ Z + , and define the function f by f ( n, q ) = q ( n + q − q ).The concavity condition is equivalent to the following inequality amongst binomialcoefficients: (cid:18) n + q − q (cid:19) ≥ (cid:18) k + q − q (cid:19) + (cid:18) n − k + q − q (cid:19) , (39) BSTRACT FACTORIALS 33 where 1 ≤ k ≤ n . We give two proofs (one analytical, and another purely combi-natorial).The first proof is by an induction argument on q . Observe that the result is truefor q = 2 as is easy to see. (Without loss of generality we assume that 2 k ≤ n throughout.) Assuming (39) true for q = m , we find that (cid:18) k + mm + 1 (cid:19) + (cid:18) n − k + mm + 1 (cid:19) = km + 1 (cid:18) k + mm (cid:19) + n − km + 1 (cid:18) n − k + mm + 1 (cid:19) , = 1 m + 1 (cid:26) ( k + m ) (cid:18) k + m − m (cid:19) +( n − k + m ) (cid:18) n − k + m − m (cid:19)(cid:27) ≤ m + 1 (cid:26) ( k + m ) (cid:18) k + m − m (cid:19) +( n − k + m ) (cid:18)(cid:18) n + m − m (cid:19) − (cid:18) k + m − m (cid:19)(cid:19)(cid:27) ≤ k − nm + 1 (cid:18) k + m − m (cid:19) + m + n − km + 1 (cid:18) m + n − m (cid:19) . (40)Using the basic identity (cid:18) m + n − m (cid:19) = m + 1 n − (cid:18) m + n − m + 1 (cid:19) in (40) and omitting the first term therein we find, upon simplification, (cid:18) k + mm + 1 (cid:19) + (cid:18) n − k + mm + 1 (cid:19) ≤ m + n − km + n (cid:18) m + nm + 1 (cid:19) ≤ (cid:18) m + nm + 1 (cid:19) , thus completing the induction argument.Another, much simpler, combinatorial argument due to my colleague Jason Gaofollows: (cid:0) n + q − q (cid:1) is the number of unordered selections (allowing repetitions) of m numbers from the set { , , . . . , n } . Next, (cid:18) k + q − q (cid:19) + (cid:18) n − k + q − q (cid:19) is the number of ways of selecting m numbers which are either all from { , , . . . , k } or all from { k + 1 , k + 2 , . . . , n } . Hence, it must be the case that (39) holds withequality holding only in degenerate cases. (cid:3) Proof. (Theorem 63) Fix q ∈ Z + . Define the function ! a as follows: 0! a = 1, n ! a = n ! qn . Clearly, n ! | n ! a for all n , while the generalized binomial coefficients (cid:18) nk (cid:19) a = (cid:18) n ! k ! (cid:19) qk · (cid:18) n !( n − k )! (cid:19) q ( n − k ) . However, both terms on the right must be integers for 1 ≤ k ≤ n since r ! | n ! for all r between 1 and n . The result follows. (cid:3) Proof. (Theorem 65) We proceed as in first part of the proof of Lemma 24, sincethe corresponding value of L = 0 here. Assume, on the contrary, that the sum P of the series is rational. Then P = a/b , for some a, b ∈ Z + , ( a, b ) = 1. In addition, P − k X m =1 p m ! = ∞ X m = k +1 p m ! . Let k ≥ b and note that p k > k for all such k ∈ Z + . We define α k as before bysetting α k = p k ! ab − k X m =1 p m ! ! . (41)Since k ≥ b and k ! divides p k ! we get that b divides p k !. Hence p k ! a/b ∈ Z + . Next,for 1 ≤ m ≤ k we have that p k ! /p m ! ∈ Z + . Thus, α k ∈ Z + , for (any) k ≥ b .As before, α k = p k ! ∞ X m = k +1 p m ! = p k ! (cid:18) p k +1 ! + 1 p k +2 ! + . . . (cid:19) . (42)Since we are in the case where L = 0 in Lemma 24 ( i.e., p k ! /p k +1 ! → α k <
1. This contradictioncompletes the proof. (cid:3) AcknowledgmentsReferences [1] N. I. Akhiezer,
The Classical Moment Problem and Some Related Questions in Analysis ,Oliver & Boyd, Edinburgh (1965).[2] M. A. Basoco,
Note on the greatest integer function , Bull. Amer. Math. Soc., (1936),720-726.[3] M. Bhargava, The factorial function and generalizations , Amer. Math. Monthly, (2000), 783-799[4] R. P. Boas,
The Stieltjes moment problem for functions of bounded variation , Bull. Amer.Math. Soc, (1939), 399-402[5] T. Carleman, Les fonctions quasi analytiques , Collection Borel, Gauthier-Villars, 1926.[6] A. M. Crabbe,
Generalized Factorial Functions and Binomial Coefficients , UndergraduateHonors Thesis, Trinity University, USA, (2001), 35 pp.[7] P. Dusart,
The k th prime is greater than k (log k + log log k − for k ≥
2, Math. Comp. (1999), 411 - 415[8] P. Erd¨os, Some problems and results on the irrationality of the sum of infinite series , J.Math. Sci. (1975), 17.[9] P. Erd¨os and R. L. Graham, Old and New Problems and Results in Combinatorial NumberTheory , L’ Enseignement Math., Universit´e de Gen`eve, (1980), 128 pp.[10] G.H.Hardy and J.E.Littlewood, Some problems of partitio numerorum, III , Acta Math., (1923), 52-54, 69. BSTRACT FACTORIALS 35 [11] G.H.Hardy and E.M.Wright,
An Introduction to the Theory of Numbers , Fourth Edition,Oxford University Press, Oxford, London, (1960)[12] Ch. Hermite,
Sur quelques cons´equences arithm´etiques des formules de la th´eorie des fonc-tions elliptiques , Acta Math., (1884), 297-330.[13] D.E.Knuth and H.S.Wilf, The power of a prime that divides a generalized binomial coeffi-cient , J. reine angew. Math., (1989), 212-219.[14] E.E.Kummer, ¨Uber die Erg¨anzungss¨atze zu den allgemeinen Reciprocit¨atsgesetzen , J. reineangew. Math., (1852), 93-146.[15] E.Landau, Elementary Number Theory , Second Edition, Chelsea Pub., New York, 1966.[16] S.Ramanujan,
Highly composite numbers , Proc. Lond. Math. Soc. (1915), 347-409[17] S.Ramanujan, Asymptotic formulae for the distribution of integers of various types , Proc.London Math. Soc., 2, XVI, (1917), 112-132[18] S.Ramanujan,
Highly composite numbers , Ramanujan J. (1997), 119-153 [Annotated byJ.-L Nicolas and G. Robin][19] J.W. Sander, On prime divisors of binomial coefficients , Bull. London Math. Soc., (1992), 140-142[20] A. S´ark¨ozy, On divisors of binomial coefficients, I , J. Number Theory (1980), 70-80[21] S.L. Segal, On π ( x + y ) ≤ π ( x ) + π ( y ), Trans. Amer. Math. Soc., (3) (1962), 523-527.[22] B. Simon, The classical moment problem as a self-adjoint finite difference operator , Ad-vances in Math. (1998), 82-203.[23] N.J.Sloane,
The Encyclopedia of Integer Sequences ∼ njas/sequences/ School of Mathematics and StatisticsCarleton University, Ottawa, Ontario, Canada, K1S [email protected]
School of Mathematics and Statistics, Carleton University, Ottawa, Ontario, Canada,K1S 5B6
E-mail address , A. B. Mingarelli:, A. B. Mingarelli: